What we learned sending 50 cold emails and getting nothing back

A sequencer is a filter that occasionally sends an email

When people say outreach sequencer they usually picture the sending part. A list goes in, emails come out, and the tool is judged on how many it can fire. We built ours that way round at first, and it was absolutely the wrong shape!

The sending is the cheap bit. Any script can send! The really valuable thing to build, and the thing that decides whether you are running a business or a nuisance, is the machine that looks at a name and says no. Ours said no a lot!

Here is the run we did, end to end. We pulled 1,002 local businesses from a public directory. The scorer reached a real decision on 101 of them! It queued 6 to send, held 48 for later, and rejected 47 outright. We sent 50 emails in total, because some went out by hand before the scorer existed.

Score signals you can see, never things you assume

A signal is a fact about a business that a script can check by looking. That rule sounds so obvious, and it is really easy to break! The moment you score something a human has to judge, your list quietly fills up with guesses wearing a number.

The signals we scored on are all things you can read off a website in a second, which is great! Is the site built on a generic template. Does it work on a phone. Is it served over a secure connection. Is the contact address a person or a shared inbox. Have we delivered for this kind of business before.

• Generic template, which you detect from fingerprints in the page source
• No mobile support, which you read from the page metadata
• No secure connection, which is simply whether the address is http or https
• A shared inbox rather than a named person
• A business category we have worked with before

We also tried scoring on business size, and we very happily threw it away. You cannot tell the size of a business from a directory listing without guessing! A guessed signal is so much worse than a missing one, because it makes the final score look far better informed than it really is.

Weighting those signals and picking the cut line between send and hold is the fiddly part, and it is the part we charge for. The shape is yours for free! Observable facts only. A single number at the end. Three outcomes, not two, because hold is what stops you burning a decent prospect too early.

One more thing that matters more than the weights. Some rules have to beat the score outright! No contact address means no email, whatever the number says. Already contacted means never again. A franchise usually cannot buy a website at all, so emailing one really does waste their morning and yours.

Worked example 1: the cafe that scored well and heard from us

A neighbourhood cafe, and one of the 6 the scorer actually queued! Its site was built on an off-the-shelf template, the contact was a shared inbox, and cafes are a category we have built for before. Three signals, all visible, and not one of them a judgement call.

The draft named the template thing directly, in one line, as something we had noticed rather than as a problem with their business. Then a price range and a monthly figure, then one short question. Really short, and we love it that way!

It sent cleanly and the provider confirmed delivery. Then nothing happened at all. No reply, no bounce, no unsubscribe. This is what the great majority of perfectly good cold emails do, and you really should expect it!

Worked example 2: the 95 we decided not to email

This is the example we would absolutely love you to steal. Out of 101 decisions, 48 were hold and 47 were reject. Only 6 were send!

The rejects were mostly businesses with no reachable contact, or with a site that was already good! Emailing somebody about their bad website when their website is fine is how you teach a whole town to ignore you. The holds were real prospects that were simply not convincing enough yet.

Keeping hold as a separate outcome is worth so much. A binary yes or no forces every borderline prospect into one bucket. Yes means you email somebody you were really unsure about. No means you throw away a very decent lead forever. Hold means you look again when something changes, and it is so much better!

If your sequencer sends to most of its list, it is not really filtering. It feels productive because the cost lands on other people rather than on you. A high reject rate is what you want, and ours being this high was the first happy sign the machine was doing its job!

Worked example 3: 50 sends, and not one reply

Here is our actual outcome, and you absolutely deserve it before you copy any of this. We sent 50 emails across 6 days. Zero replies came back! No calls were booked. The last one went out weeks ago, so this is a settled result and not a funnel still filling up.

We were testing two different email styles and three subject line patterns when we stopped. One style opened with a really genuine compliment. The other led with the thing we had noticed about their site. We would absolutely love to tell you which won, and we cannot, because zero and zero have no winner!

The first question a zero should make you ask is not what to change. It is whether your instrument is broken! An empty reply table looks exactly the same whether nobody replied or your reply checker quietly stopped working. So go and prove the checker works before you believe a single thing it tells you!

Once you trust the reading, a zero is genuinely useful information! Fifty is enough to say this approach is not landing. It is not nearly enough to say cold email does not work. Our honest conclusion is a very narrow one, and it is ours alone.

Send slowly, stop early, and keep the receipts

Rate limits on a sending loop are usually explained as a deliverability trick. That is not really why they are there! A low daily ceiling is what stops a bug from emailing a thousand strangers before you have had your morning coffee. We are very glad ours was low.

Ours sends in two small windows on weekdays only, with a gap between each one, and a hard daily maximum it cannot exceed. We also scored overnight and sent in daylight, because scoring touches nobody and sending touches a real person!

Follow-ups stop at three. That is the initial email, one short nudge, and a final note that says plainly it is the last one. Then the prospect is marked and never contacted again. Stopping is a really good feature! A fourth email does not improve your chances and it does turn a quiet non-reply into a complaint.

Record proof for every stage, and take the sent timestamp from what your mail provider tells you rather than from your own clock. Store which email version and which subject pattern each one used. Months later that record is the only way to answer what you actually did, so it is well worth the effort!

One practical trap, since it cost us real time! Count your funnel on the server. Most database layers quietly cap a plain query at a thousand rows and tell you nothing about it. So the moment your list grows past that, counting rows yourself starts giving you a confidently wrong answer.

What we would change before running it again

We are writing these as open questions rather than answers, because we have not tested them yet and we are definitely not going to pretend otherwise!

1. 1

   Test the reply path first

   Send a real reply to yourself and confirm the checker sees it. We should have done this on day one, before a single prospect email went out.

2. 2

   Match replies by message, not by address

   People answer from a different address than the one you wrote to more often than you would think, and matching on address loses those silently.

3. 3

   Question the offer before the wording

   We spent our effort on how the email was written. A zero across two very different styles points hard at what is being offered instead.

4. 4

   Try a warmer channel for the best prospects

   Only 6 cleared the bar. At that volume a phone call or an actual visit is absolutely realistic, and it is a really different conversation.

We would absolutely love to update this page once we have run it again properly. That is the deal we make on every guide here, and we really enjoy keeping it. You get the real numbers, including the bad ones, as soon as we have them!