The step where automation should slow down
Everything before this point is research. It is reversible, it is private, and a mistake costs credits. This step is different: it puts messages in front of real people with your name on them, and there is no undo.
So the build is automated and the send is not. The sequence gets constructed, the leads get loaded, and then it stops. Starting it is a decision you make, having seen what it will do.
Cadences that actually get used
LinkedIn only. A profile visit, then a connection request, then messages once the request is accepted. The right shape when you have no email addresses, and the least intrusive option for candidate approaches.
LinkedIn first with an email fallback. Same opening, but people who do not accept within a few days drop into an email touch instead. The most common multichannel shape, and it only fires the email leg for rows that actually have an address.
Email only. Three touches over about a week. Correct for business development, where work email is the right channel and volume is the point.
Linear, no branching. Fixed steps on fixed days regardless of what the recipient does. Simplest to reason about, and easiest to keep on-brand for a senior market.
What you get back
A campaign in your own sequencer with the steps and delays you chose, leads loaded, nothing running. Plus a written description of what happens on day one when you start it and at what daily volume, so the first send is not a surprise.
Where this goes wrong
Sending to an unverified list. Bounces damage the sending domain, and domain reputation takes months to rebuild. Verify before you load, especially on a list built from enrichment.
Using work email for candidate approaches. It is the wrong channel, it lands badly, and people talk about it. This is the rule the play will not bend on.
Starting the campaign to see what happens. That is what the description you get instead is for. Read it, then start.