- Four to six steps over about two weeks. Then stop.
- View profile → wait a day → connect → wait for accept → message.
- Never message within minutes of an accept. One to two days reads as human.
- A reply must stop everything queued for that person.
- If the sequence isn't working, the list is usually the problem.
Before any of the mechanics: most broken sequences are broken lists. When a campaign gets a 2% reply rate, the instinct is to add a follow-up, tighten the delays, or rewrite the opener for the fourth time. The uncomfortable diagnosis is usually that the two thousand people on the list share a job title and nothing else, and no arrangement of steps fixes that.
A good test: can you write one sentence that is specifically, painfully true for everyone on this list? If not, split it until you can. Everything below assumes you have.
The shape that works
| Step | When | What it's for |
|---|---|---|
| View profile | Day 0 | Notification. Your name becomes familiar |
| Connection request | Day 1 | Note only if you can make it specific |
| First message | 1–2 days after accept | The one that decides everything |
| Follow-up 1 | +3 days | New angle. Teach something |
| Follow-up 2 | +5 days | Specific example or proof |
| Close-out | +7 days | End it honestly. Often gets the reply |
Around fourteen days, six touches, three of which are messages. That is enough to be persistent and short enough to stay respectable. The structure is not clever, and it doesn't need to be, the leverage is in the list and the first message, not in the diagram.
Why the profile view comes first
It costs almost nothing. Profile views sit far below any meaningful rate limit, and they produce a notification, so your name has been seen once before the invite arrives. Recognition is doing real work at the moment somebody decides whether to accept from a stranger. Leave a day between the view and the request, viewing and inviting in the same second is a pattern only software produces.
Why the first message waits
The instinct is to message the moment someone accepts, while attention is high. It backfires. Nobody accepts a connection request and then watches their inbox; a message arriving forty seconds later announces that a machine was waiting. One to two days is the window where the message still has context and no longer has a timestamp that gives it away.
Branching: the part most people get wrong
A sequence is not a list. Two outcomes fork it, and handling them properly is most of the difference between outreach that reads as human and outreach that doesn't.
Accepted vs not accepted
After a connection request, people split into those who accepted and those who haven't yet. They need completely different treatment.
- Accepted: the message path. This is where your sequence actually lives.
- Not accepted: no messaging is possible, so the honest options are to wait, to view or follow them again, or to withdraw the invite after two to four weeks so it stops counting against your pending pile. Most sequences should just end here.
The delay on the first step of the not-accepted branch is what controls how long you wait for an acceptance before giving up. Two weeks is a reasonable default; acceptances after that are rare enough not to plan around.
Replied vs no reply
This is the branch that matters most, and the rule is absolute: a reply stops everything queued for that person. Not "skips the next step", stops.
A prospect who answers a question and then receives "just bumping this up" three days later has learned something specific and unflattering: nobody was reading. It is worse than never messaging them, because it retroactively reveals the earlier messages as automated too. Whatever tool you use, verify this behaviour before you launch anything, because it is the failure that costs real opportunities.
Delays, and why they should be uneven
Perfectly even spacing is a stronger automation signal than volume is. A person who messages exactly every 72 hours, at the same minute, does not exist. Two kinds of variation matter:
- Between steps: vary the gaps deliberately. 3 days, then 5, then 7 beats 3, 3, 3. It also mirrors how humans chase: keenly at first, then tapering.
- Within a day: sends should be spread across your working window with randomised gaps, not fired in a burst at 9:00. This is the tool's job, not yours, but it's worth checking your tool does it.
Working hours only, and in a timezone that makes sense for the recipient. A UK account sending at 3am is not subtle.
How long is too long
Reply rate by touch, in the rough shape everyone who has measured it reports:
| Touch | Share of total replies | Worth sending? |
|---|---|---|
| First message | Roughly half | Yes, this is the one to invest in |
| Follow-up 1 | Around a quarter | Yes |
| Follow-up 2 | Around an eighth | Usually |
| Close-out | Small but consistently positive | Yes, cheap and honest |
| Anything beyond | Marginal | No. Costs more than it returns |
The costs of a fifth and sixth follow-up don't show up in the campaign report. They show up as blocks and "I don't know this person" reports, which lower your acceptance rate, which tightens the invite ceiling LinkedIn gives that account. Over-following up quite literally reduces how much you are allowed to send next month.
Testing without fooling yourself
Test the first message. It carries most of the outcome, and it's the only step where a rewrite reliably changes the number.
- Two variants, not four. Four variants at fifty sends each teaches you nothing but costs you two hundred prospects.
- One difference between them. If A and B differ in opener, length and question, a result tells you nothing about which change mattered.
- Wait for volume. A few hundred sends per variant before concluding. At 30 sends a difference between 6% and 10% is noise.
- Measure replies, not accepts. Acceptance is decided by the invite, not by the message you're testing.
A sequence review checklist
Where sequence design meets sending limits
The two are connected in a way that's easy to miss. Your daily invite cap sets how many people enter the sequence each day; the sequence length sets how many message steps each of those people generates downstream. A sequence with three messages, fed by 20 invites a day at a 35% acceptance rate, settles at around 20 messages a day once it reaches steady state, comfortably inside normal limits.
Double the follow-ups and you double the message volume for the same number of prospects. It is worth knowing which of those two numbers you are actually short of before you add a step. What limits to set, and why covers the arithmetic, and message templates and the reasoning behind them covers what goes in each step.