Multi-touch LinkedIn outreach: the pre-launch checklist

A multi-touch LinkedIn sequence is a connection request followed by two or three messages spread over a couple of weeks. Done well, it turns a cold profile into a conversation. Done badly, it is the same pitch three times, sent to someone who already said no, from an account that LinkedIn is about to restrict.

I posted a short version of this on X on September 5 with Aditya, mostly as a warning: do not open a connection request with a sales pitch. This is the full checklist we run at PumpGTM before any sequence goes live for a customer, and it works the same if you are sending by hand.

Step 1: Give every touch one job

Write down what the recipient should understand after each interaction. Not what you want to say, what they should take away.

A sequence that works usually looks like this:

1. Connection request. One line about who you are, one question about whether they have the problem. Nothing else. The job of this step is to get accepted.
2. First message, after acceptance. Thank them, then ask the question again with one specific detail from their profile or their company. The job is to get a yes or a no on the problem.
3. Second message, three to five days later, only if silent. One useful thing: a short observation, a relevant example, a link that answers a question people in their role ask. The job is to give a reason to reply that is not "buy from me".
4. Last message, a week later, only if still silent. Close the loop politely and leave the door open. The job is to make it easy to come back later.

If two steps carry the same pitch, delete one. More touches do not make an irrelevant message relevant. They make it annoying.

Keep the "why this person" note next to each name. The sequence should reflect their company and the problem you solve for it, not just their job title. This is the same rule as the "relevant because" column in how we found customers for Supermemory.

Step 2: Decide what counts as a reply, and what it stops

This is where most sequences hurt you. Before enabling anything, check:

• How are replies detected? Does a one-word answer count? Does an emoji reaction?
• Which scheduled steps does a reply cancel? All of them, or only the next one?
• What happens to a reply that lands between two scheduled steps?

The rule is simple: a person who answers gets a human response that acknowledges what they said, and every scheduled follow-up for them stops. If someone asks you to stop, record it and stop. No exceptions, no "one last message".

PumpGTM pauses the sequence on any reply and shows you the thread so you decide what happens next. If your tool cannot show you that behaviour in a test, do not run it on real prospects.

Step 3: Treat account limits as a separate check

LinkedIn caps connection requests at roughly 200 a week and watches for behaviour that looks automated. A pacing setting in a tool is not a guarantee that the account stays unrestricted.

Before running a sequence:

• Look at the account itself. Any warnings, any "we've restricted..." notices, any unusual acceptance drop.
• Check LinkedIn's current rules, because they change.
• Spread requests across the week rather than sending the weekly cap in one sitting. Steady looks human.

If the account shows a restriction, stop and investigate. Whether the account can take an action and whether the action is appropriate are two different questions, and both need a yes.

Step 4: Keep the walkthrough honest

If you record a video to explain the sequence to your team or a customer, show the exact step the text describes. Label demonstration data as demonstration data. A count in a sample interface is not a customer result, and the moment you present it as one you lose the trust the whole sequence depends on.

Step 5: Measure conversations, not activity

Four numbers, tracked separately, for a fixed date range:

Stage	What to count
Invitations sent	Every connection request in the period
Accepted	Requests accepted in the period
Replies	People who wrote back, any content
Qualified conversations	Replies where the person confirmed the problem or asked a real question

Use accepted connections as the denominator for reply rate. Record the sequence version alongside its source list and campaign. When you change copy or timing, keep the old version so you can explain what changed and why.

Then read the replies, not just the count. "Not my team" is targeting feedback. "How does this compare to X" is a qualified conversation. A polite no is neither. This is the same discipline as reviewing an AI-written cold message: the words in the reply tell you what to fix.

Questions we get about multi-touch sequences

How many touches is too many? Beyond four, replies almost never improve and complaints do. Three messages after the connection request is the practical ceiling.

Should the connection request mention the product? No. Ask about the problem. The product comes up when they say yes.

How long between touches? Three to five days, then a week. Faster reads as pressure, slower and they forget who you are.

Can I run the same sequence for every prospect? You can, and it will underperform. At minimum the first message after acceptance needs one specific detail per person.

Do I need a tool for this? Not at first. A spreadsheet and a calendar work for twenty prospects. Past that, the reply-handling rules in step 2 are where people slip, and that is the part PumpGTM automates from your own LinkedIn account. Start at pumpgtm.com and you approve the list and the messages before anything is sent.