Quotes that never got a yes or a no. Invoices past due. A customer who booked and did not turn up. Chasing all three is repetitive, unpleasant, and automates well, right up to the moment somebody actually replies.
A follow-up sequence is defined by the condition that starts it and the condition that stops it, and both have to come from real data. Quote sent more than five business days ago with no linked response. Invoice past its due date with no payment recorded. Booking marked as a no-show. If your systems cannot answer those questions reliably, the sequence cannot be built reliably either, and that is the first thing worth checking. Most failed follow-up automations failed at the trigger, not the sending.
The list of reasons not to send is longer than the list of reasons to send, and every one of them has to be wired in: the customer already replied, already paid, is in a dispute, has unsubscribed, is being handled by a specific person, or is a customer you would never chase this way. A sequence that sends a payment reminder to someone who paid yesterday does more damage in one message than the automation saves in a month. This is the part that takes the time to build, and it is the part worth paying for.
The automation should stop the moment a human replies, and it should be genuinely difficult for it not to. This is why the calculator caps the automatable share for follow-up at 90 percent rather than higher: the sending, the scheduling and the tracking go, and the conversation that starts when someone answers stays with a person. Software that tries to carry that conversation is a different product with a different risk profile, and it is not what a chasing sequence should be doing.
The common objection is that automated chasing feels impersonal. In practice a short, plainly written, correctly timed reminder reads better than the version a busy person sends at 9pm, because it is consistent and it is not annoyed. What does not survive is volume: a sequence of six messages over two weeks reads as pressure regardless of how it is written. Two or three, spaced properly, with a real stop condition, is the shape that works.
Whether the trigger data exists and is trustworthy. What your current follow-up actually is, honestly, because for many businesses the answer is that it does not happen at all, and the comparison is not automation against a person, it is automation against nothing. And who owns the replies once they start arriving, because a sequence that generates responses nobody answers is worse than no sequence.