Almost every argument about cold email sequences is an argument about length, and length is the least interesting variable. A four step sequence where every step earns its place outperforms a nine step sequence where six of the steps are a person typing "thoughts?" into a scheduler, and it does so while generating a fraction of the complaints.
This is about what each step is for, what it is allowed to assert, when to stop, and the specific patterns that turn a sequence from a pipeline source into a deliverability incident.
The shape of a sequence that works
A cold sequence is not a campaign that repeats. It is one argument, delivered in instalments, where each instalment adds something the previous one did not contain. If you cannot say what step three adds, step three should not exist.
The structure that holds up across markets looks like this:
- Step one: the observation and the ask. Something specific and true about them, the reason it makes you relevant, and one small request.
- Step two: the different angle. The same problem approached from another direction, because the first framing may simply not be their problem.
- Step three: the evidence. Something concrete: how the thing works, what a comparable situation looked like, a number you can stand behind.
- Step four: the close. A short, unembarrassed acknowledgement that this is the last one, with a genuinely easy way to say no.
Four instalments, four different contents. Notice that none of them is "checking in", because checking in is not an instalment, it is an interruption with no payload.
How many steps, and how far apart
Between three and five for a cold sequence. Below three you are abandoning threads that were merely unread rather than unwanted. Above five, in cold outbound, you are usually padding, and the marginal step is contributing complaints rather than replies.
On spacing, the useful principle is that gaps should widen, not stay even. Something like three days, then four, then seven, then ten. Even spacing reads as a machine. Widening gaps read as a person who is interested but not desperate, and they give a busy prospect a realistic chance to come back to the thread.
Send inside working hours in the recipient's timezone. Not because of a magic hour, the studies on that contradict each other and most of them are vendor marketing, but because an email that arrives at three in the morning local time is visibly automated and gets treated accordingly.
What each step is allowed to say
Here is the rule that matters more than any of the structural advice: every factual claim in every step must be traceable to something you can open. Not just in step one. In all of them.
This is where sequences go wrong in a way that single emails do not. Step one usually gets researched, because a human wrote it or reviewed it. Steps two through four get generated from a template, and the template contains slots, and a model filling a slot on step three has less to work with than it had on step one, so it reaches for the invented consensus: "most teams in your position find that...". Nobody reviews step three. Step three goes out four days later, to somebody who did not reply to step one, which means they were already unconvinced.
A practical rule: if a follow up cannot say something specific and sourced, it should ask a question instead of asserting something. A question is not a weaker email. It is the only kind of email that has a reason for the reader to press reply.
Why the bump that says nothing gets complaints
The one line bump, "any thoughts?" or "bumping this to the top of your inbox", is popular because it is free to write and it does produce some replies. It is also the single most reliable complaint generator in cold outbound, and the reason is worth understanding.
A complaint is not a reaction to volume. It is a reaction to being treated as a resource. An email with content, even unwanted content, is a person making a case. An email with no content is a person poking you to see if you are awake. The second one produces the specific irritation that makes somebody reach for the spam button rather than simply deleting it, and the spam button is the only reaction that costs you anything at the domain level.
If a step cannot carry content, delete the step. A three step sequence with three real messages outperforms a six step sequence with three real messages and three pokes, on replies as well as on complaints.
The stop conditions
A sequence needs to stop on all of these, automatically, without anybody remembering:
- Any reply. Including an out of office in most cases, at least until the date it gives, and certainly including a reply that says no.
- Any opt out, in any form. The word stop in a reply counts. A person should never have to say it twice, and the ten business day ceiling in the law is not the target.
- A hard bounce. Immediately, on the address and on the sequence.
- Somebody else at the account replying. If a colleague answered, the thread is handled.
- A rising complaint rate. The whole sequence, not just the one contact. If complaints are moving, the message is the problem, and continuing to send it is the most expensive possible way to find that out.
How a sequence burns a domain
It is rarely one dramatic mistake. The usual pattern is four ordinary decisions compounding.
- A list is bought or scraped, so the addresses were never expecting anything from you.
- The sequence is long, because more steps produced more meetings in a spreadsheet somewhere.
- The later steps have no content, because nobody wrote them properly.
- Volume ramps quickly, because the pipeline number needs to move this quarter.
Each of those alone is survivable. Together they produce a complaint rate above the threshold the mailbox providers publish, and at that point the consequence is not a fine or a warning. Delivery degrades for everything sent from that domain, including invoices and password resets, and recovering takes months of low volume good behaviour.
Google's sender guidance is explicit about the number to watch: keep the spam rate reported in Postmaster Tools below 0.30%, and aim to stay under 0.10%. Those thresholds, along with SPF, DKIM, DMARC and one click unsubscribe, have applied to bulk senders since February 2024.
Volume, and the number that actually matters
Most teams manage volume by deciding how many emails they want to send. The better approach is to decide what complaint rate you are willing to run at, and let that set the volume.
In practice that means a daily ceiling per mailbox that you set deliberately and do not raise casually. Start low, watch the rate in Postmaster Tools, and raise the ceiling only when several weeks of data support it. A ceiling that the sending tool itself cannot exceed is worth a great deal more than a throttle somebody can raise on a Thursday afternoon when the quarter is behind, which is precisely when it will be raised.
This is why our email sequence software treats the cap as a structural limit rather than a setting, and why each follow up is written against the thread rather than from a template with a merge field in it.
A four step sequence, written out
Not a template to copy, because a copied template is exactly the thing that stops working. A shape, with the reasoning attached.
Step one, day zero. One specific observation about them, sourced. One sentence connecting it to a problem you address. One question. Under a hundred words. No attachment, one link at most, and a real signature with a postal address.
Step two, day three. A different angle on the same problem, not a repetition of the first. If step one was about cost, step two is about time, or risk, or the person who currently does it manually. Explicitly not a "just following up", because you are not following up, you are saying something new.
Step three, day seven. The concrete one. How it actually works, what changes in the first month, what a comparable situation looked like. If you do not have a real reference you can describe without embellishing it, describe the mechanism instead. A specific mechanism beats a vague result.
Step four, day fourteen. Two sentences. This is the last one, here is the one thing to reply with if it is ever relevant, and nothing more. No guilt, no "I will assume you are not interested", no final offer. The dignified close is also the one that most often gets a reply, because it is the only message in the sequence that costs the reader nothing to answer.
Then stop. A sequence that stops when it says it will is the only kind that leaves you able to write to that person again in six months, which is where a surprising share of pipeline actually comes from. If you want the mechanical checks that keep each of those four steps honest, they are set out on the cold email software page.