I Sent 4,535 Cold Emails to Startup Founders. Ten Replied.

Between 14 January and 4 August 2026 I sent 4,535 emails to startup founders. Ten of them wrote back.

That is not a deliverability failure. 3,004 of those addresses were verified before a single send, and the campaign bounced at 3.7% — 170 hard bounces out of 4,535. The mail arrived, in the inbox, at a rate any deliverability consultant would sign off on. Founders read it and did nothing.

I'm writing this from the builder's seat, not the marketer's. The outreach platform is mine — I built it and I run it, and how it works stays in-house. What I'm publishing is its funnel, including the numbers that make my own channel choices look bad.

One definition first, because everything hangs on it. A reply here means the lead wrote back: a conversation opened. That is where I stop the funnel, on purpose. What happens after a founder answers is between them and us.

9 out of 10 people I wrote to were a founder, a CEO or a CTO

The sample is the credibility, so all of it: 4,969 leads across 4,360 companies in 37 countries, 4,972 contacts. Of the 4,968 contacts with a title recorded, 90.6% are founder, CEO or CTO — this is not a database of marketing managers standing in for decision-makers.

Where the list came from stays in-house; it's enough to say we look for founders, scale-ups and companies in our own sector, and that the list is narrow by design. By profile, 73% are digital SMEs, 17% SaaS scale-ups, 8% solo founders — the segment I wrote about in The Advent of the Solo Founder — and about 1.5% fintech. One split I'll need later: 333 of the leads came from my own LinkedIn network; everything else was cold.

On top of that list the platform has logged 11,009 touches all-time, 10,489 of them in 2026 across 140 active sending days: about 75 touches a day. Two channels. On LinkedIn, 2,127 connection requests, 2,832 messages and 49 InMails. Over email, 4,535 sends in a three-step sequence.

3,021 of those leads were actually contacted. They produced 274 replies.

Cold email to founders returns two replies per thousand sends

4,535 emails. 10 replies. 0.2%.

Verified addresses, a 3.7% bounce rate, three steps, a subject line and body tailored to each company. Two answers per thousand sends.

Two replies per thousand emails is not a copywriting problem. It is a channel that has been closed for a while, and nobody sent a notice.

Be precise about what that number covers. It says cold email to founders, from a sender they have never heard of, does not open conversations at a rate that justifies the infrastructure. It says nothing about email between people who already know each other, about newsletters, or about your inbound. My ten replies were real conversations with real founders. Ten.

96 of every 100 replies arrived on LinkedIn

Split the replies by channel and the asymmetry is the whole story: 266 LinkedIn reply events against 10 email replies (events, not leads — a few founders wrote back more than once). About 96 of every 100 conversations I opened this year started on LinkedIn.

The gate that matters sits earlier. Of 2,123 connection requests, 587 were accepted — 27.6%, a bit more than one in four. Nothing downstream recovers a rejected request, which makes the request itself, and not the message that follows it, the piece of copy worth an afternoon.

Why a founder answers on LinkedIn and ignores the inbox

My data shows the asymmetry; it doesn't explain it. Six explanations I'd defend, from building the machine and reading every reply that came back:

1. A founder's email address is public infrastructure. It's on the deck, in the domain registration, in every SaaS signup they ever made. It receives vendor mail as a category, and it gets triaged as a category — often by someone who isn't the founder. The LinkedIn queue is smaller and hasn't been delegated yet.
2. LinkedIn asks for a cheap yes first. A connection request costs one tap and commits to nothing; 27.6% granted it. Email opens by asking a stranger for the expensive thing — a considered written answer — on first contact.
3. Identity is verifiable at a glance. Face, employment history, mutual connections, a company page one click away. A from: header proves nothing, and founders know it.
4. One line is a complete reply on LinkedIn. «Not now, ping me in Q4» reads as a normal message there. The same sentence as an email reply reads as a brush-off, so the founder saves it for when there's time to write properly. That moment doesn't come.
5. The thread carries its own history. My second and third LinkedIn touches sit under the first, one screen, one conversation. A third email to someone who ignored the first two is a stranger writing for the third time.
6. The notification lands where the founder already is — the phone, in the gap between two meetings. Email is a desk task competing with other desk tasks.

None of that is a law of nature. It describes where founders currently keep their attention — and that is where 96 of every 100 replies reached me.

Spanish founders replied at nine times the US rate

The channel gap is large. The geography gap is larger, and it surprised me more.

Reply rate by market, over the 3,021 contacted leads:

• Spain: 25.6% — 193 replies from 754 leads. One in four.
• United States: 2.9% — 22 replies from 752 leads. One in thirty-four.
• Rest of world: 3.9% — 59 replies from 1,515 leads.

Two nearly identical volumes, 754 and 752, in two markets, with the same platform, the same sequences and the same ICP. Spanish founders replied at roughly nine times the US rate. If you run one blended reply rate across a multi-market campaign, that is the size of the thing your average is hiding.

Warmth beat every channel decision I made

And then the number that outranks all of the above. Leads sourced from my own LinkedIn network replied at 42.3% — 141 from 333 contacted. Everything cold replied at 4.9% — 133 from 2,688.

8.6×. 333 warm leads produced more conversations than 2,688 cold ones. No subject line, no send time and no channel switch in this dataset came close to that lever. Channel is a routing decision; warmth is the input.

The queries that bring us visits disagree with my funnel

The outbound log is only half the instrumentation. Conectia also has a Search Console, and while the machine was pushing, search was pulling in the other direction. I'll give shares only, no volumes — the search dataset is young, and the shape is the point.

Three of every four search visits to conectia.pro land on the blog, not on a commercial page. Of the visits search can attribute to a query, about half are our own name typed into the box, and brand queries are roughly a quarter of everything we get shown for. That's the warmth finding again in different clothes: the people who arrive are the people who already knew we existed.

The discrepancies start when you compare what I do with what people look for. Queries about outreach or lead generation — the thing I spent 140 sending days on — are under 1% of our impressions and have produced exactly zero visits. What does surface us, about one impression in five, is the other side of the trade: how to find senior engineers, alternatives to the big talent marketplaces, what a nearshore partner should look like.

And then the one I didn't expect. Roughly one in twelve of the distinct queries that surface us are not keywords at all — they are whole paragraphs, questions a founder typed to an AI assistant that went searching on their behalf. One of them, verbatim: «I'm a technical founder, we have three enterprise customers and revenue coming in, but it's just me building everything. How do I actually find that person?» That is the person I emailed 4,535 times. These assistant-shaped questions are about 4% of our impressions and, so far, 0% of our visits: the founder asks the machine, the machine reads us, the founder never clicks.

Both datasets describe one behaviour. Cold email returned two replies per thousand because founders no longer answer strangers in the inbox. The search log shows where that attention went: they ask their own tools, in full sentences, about exactly the problem my emails offered to solve. The conversation I kept trying to start is already happening — one layer up, without me in the room.

What this dataset can't tell you

The honest limits, because the first thing a skeptic does is go looking for them.

The channel comparison isn't like-for-like. LinkedIn requires acceptance before a message, so those 266 reply events come from an audience that already said yes once. Email has no such gate. That's a real objection, and the raw pair is still lopsided: 2,123 requests plus 2,832 messages produced 266 reply events; 4,535 emails produced 10.

Spain and warmth are tangled and I can't fully separate them. My network is largely Spanish, I sell in that market, and I write to Spanish founders in their language and mine. Both effects live inside the same 274 replies. Read the 9× as Spain-plus-me, not as a property of Spanish founders.

n = 274. Enough to see a 9× gap and an 8.6× gap. Not enough to conclude anything about a subject line, a send hour or a sequence length.

This is observational, not an experiment. Copy, targeting and sequencing all changed over 140 days. Nothing here was randomized.

The search shares come from a small, early sample. That's why they are percentages with no volumes attached. Read them as direction — the asymmetries are large enough to survive the noise; the decimals are not.

And it's vendor data. Conectia sells nearshore engineering squads — deployed, not staffed — and every lead in this dataset is someone I hoped would become a client. The figures are exact and the funnel is one company's.

What I'd do if I were running founder outreach this quarter

1. Make LinkedIn the primary channel and email the follow-up surface. Budget the hours the way the replies actually arrive, not the way the tooling market is priced.
2. Optimize the connection request, not the message. 27.6% is the gate the entire funnel passes through. A rejected request is a lead you cannot reach again this year.
3. Exhaust your own network before you buy a list. 333 warm leads beat 2,688 cold ones outright. Map the second-degree connections you already have before paying for first contact with strangers.
4. Segment by geography before you segment by ICP. Report Spain and the US as two campaigns. A single blended number would have told me my outreach was mediocre everywhere, when it was 25.6% in one place.
5. Keep verifying the addresses anyway. A 3.7% bounce rate is what keeps the sending domain alive, and those ten conversations were worth having. Just don't fund the channel as if it were the engine.
6. Instrument before you scale. Every touch and every reply logged and joined back to its source — that record is the only reason any of this post exists. It cost an afternoon in January.
7. Answer the questions founders ask their machines. The founder who ignores your cold email is typing their problem, in full sentences, into an assistant. A published answer is the only touch that reaches that surface — and it's the one channel in this whole post where the buyer starts the conversation.

The funnel in seven numbers

• 10,489 touches in 2026 over 140 active sending days — about 75 a day.
• 90.6% of contacts were a founder, CEO or CTO.
• 4,535 cold emails produced 10 replies — 0.2%, at a 3.7% bounce rate.
• 266 LinkedIn reply events against 10 by email: 96 of every 100 conversations.
• 587 of 2,123 connection requests accepted — 27.6%.
• Spain 25.6% against the US 2.9%: one in four, against one in thirty-four.
• Own network 42.3% against cold 4.9% — 8.6×.

The machine turned out to be good at the part I assumed was hard: finding 4,969 leads in 37 countries, verifying them, and writing to each one about their own company. The hard part sat upstream of all of it — who already knows you, and where they are. Two replies per thousand emails is what a stranger gets. 42.3% is what someone who has met you gets. Everything I build into the platform next is aimed at making the second number carry the volume.

---

If you're building this kind of machine in-house, it's ordinary software with a model in the middle, and it needs engineers who've run systems in production. Talk to a CTO about the squad that builds it.