Cold email not working? Check in this order
Cold email stops working for one of seven reasons, and the order you check them in matters more than the list, because three of those reasons can make everything you check next untrue. Below: the order, the observation that settles each branch, and what checking out of order costs you.
By Rahul Bageria, co-founder · Updated August 2026 · 11 min read
The order, and what settles each branch
Seven things can be wrong, and three of them corrupt the reading of the other four. Those three go first. The table is the method; the reasoning follows.
| Check | What you look at | What settles it |
|---|---|---|
| 1. Volume and elapsed time | Contacts reached, and days since the last send | Under 150 contacted, or under three days |
| 2. Did it land | One message header, and replies per sending address | One mailbox quiet while its siblings hold |
| 3. The list and the data | Bounce rate, and who the misroutes name | Misroutes repeating across different accounts |
| 4. The offer and the ask | The wording of any refusals you got | A no that explains itself |
| 5. The copy | The first line and the ask | Nothing, until the four above are closed |
| 6. The replies you already have | Your last fifty replies, sorted | Anything worth a calendar invite |
| 7. The number | What your tool counts as a reply | One address replying twice, counted twice |
Sources: the contact, reply and bounce figures here are vendor data from Smartlead's State of Cold Email 2026, built from more than 850 million emails sent through its own platform from January to June 2026, read August 2026. The 150-contact and three-day lines are gates we set from it.
Why you do not start with the likeliest cause
Every list on this topic is sorted by how common each cause is. That is the wrong sort. A cause goes first because it can make the next check untrue.
Evidence order ranks a check by what it can invalidate. Likelihood order ranks it by how often that cause is the answer. Likelihood is the right sort inside a branch and useless across branches, because one branch corrupts the readings of the others.
- •Not enough sent, or not enough time: every number is noise
- •Mail that did not land: every message test is a false negative
- •A list of the wrong people: a good offer looks unwanted
- •The offer, the ask, the copy, your filed replies, the number
- •Each one reads true once the three above are closed
- •Guess wrong here and it costs a rewrite, not the evidence
Nobody reads inbox placement directly, including the platforms: Smartlead uses replies as a stand-in because it cannot see where its own senders landed. So the test is a header and a comparison between your mailboxes, and the only published thresholds, traced in our cold email deliverability guide, are one provider's account policy.
You are not looking for the likeliest cause. You are looking for the one that would make everything you check next a lie.
One stalled campaign, walked down the ladder
The hard case: 900 people contacted over four weeks, three emails each, and not one reply to read. Five branches close in twenty minutes.
An illustrative walkthrough of the method, not a specific client result. We report real numbers only when they are real.
-
1
Count what you sent
Nine hundred clears the gate. The typical sender in Smartlead's benchmark needs about 135 contacts per reply, so 900 should have produced six or seven. Zero is a finding.
-
2
Read the clock
The last send went out nine days ago. Replies arrive fast: 64 percent inside the first hour and 94 percent within three days, across 7.9 million of them. More waiting buys nothing.
-
3
Read one header
Send one message from the campaign mailbox to an inbox you own and check that SPF, DKIM and DMARC passed. Those three DNS records let a receiving server confirm the mail is really yours. All three pass.
-
4
Read the bounce rate
It sits at 1.1 percent, as the platform counts a bounce. The typical sender is near 1.5 and the upper quarter at 3.2 and above. Not the loud kind of broken.
-
5
Compare the mailboxes
Four sending addresses, all flat at zero. One quiet mailbox beside three healthy ones is placement. Four quiet ones point past delivery, to the list or the message.
-
6
Reply to yourself
Two replies from one test address move the counter by two, so the tool counts what you assumed. The campaign really did produce nothing.
Twenty minutes, five branches closed, no reply needed. What is left is the fork no amount of staring settles: the wrong people got it, or the right people had no reason to care.
We check in this order because we have paid for the other order.
On a stalled account we open the send count and one header, not the copy doc. Rewriting feels like progress and spends your best version on a test nobody could read.
The seven branches, and the test in each
One row per branch. The test is this page's job; the fix belongs to the page each row names.
Have you run the test yet?
The typical sender in Smartlead's benchmark needs about 135 contacts per reply, and plans about 1,350 for ten replies. Sixty emails on a Tuesday is not a result. The evidence ladder for volume sets what each number buys.
One mailbox quiet, the others fine
Count replies per sending address, not per campaign: one address flat while its siblings hold is placement. Likelihood order is the right sort inside this branch, and the nine causes are ranked in why cold emails go to spam.
The misroutes name the seat
Misroutes repeating across accounts indict the seat you picked; two desks at one account both saying it does not apply indict the account. Sourcing and hygiene are in lead list building.
A no that explains itself
Silence does not settle this branch and neither does a rewrite. What settles it is a refusal that names its reason: already solved, or nobody owns the problem. Where offers break is the outbound offer.
The branch with no test of its own
No first line can be judged while the four checks above are open. That is why it sits fifth, not because writing matters less. The craft is in cold email copywriting.
Your pipeline may be in your inbox
A misroute naming the right person, a not-now with a month attached, a different problem offered instead. Sort your last fifty replies against a rule you wrote down. The workflow is reply management.
The campaign may be fine, the number wrong
Some tools divide replies by emails sent, so a longer sequence lowers the fraction with nothing changing. Reply twice from one address and watch whether it moves once or twice. Definitions live in outbound metrics.
What checking out of order costs you
Checking out of order does not cost a week. It costs the evidence, which is the one thing you cannot buy back.
"It is probably the copy. Rewrite it tonight, send again Monday, and see."
- ✕Spends the best version you had on an unreadable test
- ✕Returns a false negative if the mail never landed
- ✕Ends with a founder concluding the market said no
"Close the free branches first, then change exactly one thing."
- ✓The free checks are the upstream ones, which is the lucky part
- ✓Evidence order and cost order agree on this problem
- ✓One change a week, so the next result means something
Rewriting the copy first
The rewrite is the default because it is the only branch you can act on this afternoon. The cost is not the afternoon: your best angle is spent on a segment you cannot read again.
Buying a tool first
A new sequencer changes who sends, not whether anyone wanted it. You have added a variable to a test that had too many, plus a month before the next reading.
Widening the list first
More contacts on a corrupted reading make a bigger corrupted reading. Volume is a gate you pass once, not a fix you apply twice.
Changing five things at once
Whatever happens next, nobody can say which change did it. Two people will argue about it for a month, both reasoning from the same unreadable result.
Delivery does not go first because it is usually the problem. It goes first because when it is the problem, every other answer you get is wrong.
Want your stalled campaign read in the right order?
Book a Fit CheckWrite the diagnosis down in one line
A diagnosis is not finished until it fits in one line you could send to the other person on the team.
Write down the conclusion
Cold email is not working. Trying new copy and a new tool this week.
- ✕No counts, so nobody can check it later
- ✕Two changes, so the next result is unreadable
- ✕Cannot be tested, only argued about
Write down the reading
16 Aug: 900 contacted, 9 days quiet, header passes, bounce 1.1%, all four mailboxes flat. Not delivery. Next: one segment, 400 contacts, a different angle.
- ✓Counts first, so the next one is cheaper
- ✓Names the branch ruled out, not just the plan
- ✓One change, with a number to read it against
Twenty minutes, and one change a week
Two people, no dashboard, a timer. Everything upstream in the order is free and takes minutes. Everything downstream costs days.
- 1 Three checks can invalidate the rest: volume, delivery, list.
- 2 Zero replies under 150 contacted is an unfinished test.
- 3 The cheap checks are the upstream ones, which is the lucky part.
- 4 One branch, one change, one week. Then read it.
If the bounce rate is what moved, verify at send time rather than build time; the email verification tools we have reviewed differ mostly in when they check.
The twenty minutes is easy. Doing it after a bad week is not.
Nobody follows an order while they are frightened of the answer, and that is when five things change at once. Put the checks in a calendar block with two people in it.
Four ways a diagnosis goes wrong
Every bad diagnosis is one of four, and three come from reading a number that could not carry the weight.
Images load themselves now, which is why Smartlead leaves open rate out of its report: reported opens of 60 to 70 percent are largely automatic. Treat opens as unknown data.
Three misroutes feel like a pattern and are not one. Which layer is broken, the account or the seat, is worked through in ICP versus buyer persona.
An out-of-office and a warm referral do not belong in one number. What counts as positive, and the gates it passes, are in what a positive reply rate measures.
Nobody wants this is the one conclusion the test cannot support until every branch above it is closed. It is also the only one you cannot undo.
Starting over is not a diagnosis
New domains, a new tool and a new list in the same week is the most expensive move on this page. You have thrown away the only reading you had, added weeks of warmup, and you still do not know which branch was yours.
Questions founders ask
Why am I not getting any replies to my cold emails?
How many cold emails do I need to send before no replies means something?
How long should I wait before deciding a campaign is not working?
Is it my copy or my deliverability?
Will fixing SPF, DKIM and DMARC increase my replies?
Should I remove the unsubscribe link to get more replies?
Co-founder of Real Good GTM. He has been the first business hire and Chief of Staff at seed-stage B2B startups, building outbound pipeline before any playbook existed. This post is the order he runs on a stalled account before anyone is allowed to open the copy doc, and the reason the free checks come first.
Connect on LinkedInOnce you know which branch is yours
The three branches that turn out to be the answer most often, in full.
Why cold emails go to spam
The nine causes of poor placement, ranked by how likely each is yours, each with a ten-minute confirm.
Read the postThe outbound offer
What you are actually asking for, why most offers break, and how a no tells you which part broke.
Read the guideReply management
Sorting, answering and converting replies, including the ones a stalled campaign already produced.
Read the guideWant the diagnosis run properly, by the two of us?
Book a fit check. We'll walk your last campaign down this order, tell you which branch is yours, and tell you straight if outbound is not the right motion for you yet.
Book a Fit CheckNo hard sell. No fake numbers. Real good work speaks for itself.