Why the second follow-up outperforms the fifth

The reply rate of each touch, counted on cold threads

One accounting rule decides this whole table, so it comes first. A touch counts as cold only when the thread was silent before it went out. Once a person replies, every message we send afterward belongs to a live conversation, and a live conversation answers on its own momentum. Mixing those two populations is the single easiest way to convince yourself that touch six works.

With that rule applied, here is what each touch returned over the window:

| Touch | Cold touches sent | First replies | Reply rate |

|---|---|---|---|

| 1 | 1,331 | 160 | 12.0% |

| 2 | 749 | 39 | 5.2% |

| 3 | 506 | 29 | 5.7% |

| 4 | 75 | 1 | 1.3% |

| 5 | 13 | 1 | 7.7% |

| 6 | 1 | 0 | 0% |

Read the cumulative line instead of the rows and the shape gets sharper. Of the 1,331 threads that received a first message, 160 answered it, which is 12.0%. Carrying those threads through a second and third touch brings the total to 228 replies, 17.1% of threads messaged. Carrying them further brings it to 230, 17.3%. The second and third touches together add 5.1 points of reply rate. The fourth and fifth add 0.2 points across 89 attempts.

Put the same thing as a share of outcomes: 228 of 230 first replies, 99.1%, arrived on one of the first three touches. Follow-ups earned 70 of the 230, so 30.4% of everything we got came after the opener. That is the case for following up at all, and it is a strong one. It is also the case for stopping, because the same log shows where the curve flattens.

Two honest limits on the table. Touch five shows 7.7% on thirteen attempts, which is one reply and a wide interval, so the raw count is the number to read there. And the fourth and fifth rows are thin because our own policy caps the sequence, so the deep end of the table measures a cadence we deliberately keep short. The method for reading small samples honestly is in A/B testing outbound copy when the sample is small, and it applies to your own sequence table as much as to copy tests.

The follow-up replies came in at the same quality

A falling reply rate would be easy to accept if the replies that follow-ups produced were worse: polite brush-offs and wrong-person redirects, the kind of answer that costs an hour and ends there. So we checked the grade of the reply against the touch that earned it.

Of the 160 people who replied to a first message, 10 now sit at a qualified conversation or a booked call, 6.3%. Of the 39 who replied to a first follow-up, 3 do, 7.7%. Of the 29 who replied to a second follow-up, 2 do, 6.9%. Three samples, three rates inside a point and a half of each other, on numbers too small to separate.

So the finding points one way. Touches two and three return fewer replies at the same grade, and those replies are worth having, which is why a sequence that stops at the opener leaves money on the table. Quality holds flat as depth grows, so the case for touch six has to come from volume, and volume is where the table says it loses. Our full definitions of a qualified conversation and a booked call, and what each one costs us, are in cost per meeting from LinkedIn outreach.

A touch is answered inside two days, or it waits for the next one

Timing explains why the curve falls the way it does. Of the 230 first replies, 197 arrived within 48 hours of the touch that earned them, 85.7%. By touch: 138 of 160 on the opener, 36 of 39 on the first follow-up, 21 of 29 on the second. The median gap between a touch and the reply it earned was the same day for touches one and two, and 0.8 days for touch three.

A person who answers a given message answers it almost immediately, because the answer depends on the week it arrived in. Sending the next touch three days later puts the same question into the same week and gets the same silence back. Sending it ten days later puts it into a different one.

The same clock runs on our side of the thread. A reply is live for about as long as it took to arrive, so an answer inside the same working day is what keeps it alive. We hold a two-day ceiling on every inbound message and route the safe classes automatically, and what to put in that first answer is in what to send after the first reply to a cold message.

Spacing carries more weight than count

If the week decides the reply, the interval between touches should matter more than their number. It does. Grouping every follow-up by the gap behind it:

| Gap behind the touch | Touches sent | Replies | Reply rate |

|---|---|---|---|

| Under 3 days | 62 | 2 | 3.2% |

| 3 to 7 days | 667 | 33 | 4.9% |

| 7 to 21 days | 528 | 32 | 6.1% |

| 21 days or more | 87 | 3 | 3.4% |

The middle band wins. A follow-up sent 7 to 21 days after the last one answered at 6.1%, against 4.9% for the common 3 to 7 day spacing and 3.2% for anything under three days. The 62 touches inside three days produced two replies between them, which is the clearest result in the table even on a small sample: a question re-asked within 72 hours is the same question.

The far end falls off too. After 21 days the rate drops to 3.4%, and the thread needs a reason for its own existence by then. Our own medians land close to the middle band by design: 4.0 days from the first message to the second, 10.0 days from the second to the third.

Three touches with real air between them span about two weeks, and that is the window where a thread is worth the room it takes in your sending budget. Budget is a real constraint here, because the platform caps invitations at a weekly ceiling that binds long before your cadence does, so every touch spent on an old thread is a first touch you skipped.

The cadence we run, and how to set yours tomorrow

The policy that came out of these numbers has four parts.

The sequence is three touches: an opener, a follow-up at roughly four days, a second follow-up at roughly ten. Each one carries a new point, because a restatement spends a touch on ground already covered. Our text checks run on every outbound message before it leaves, 134,762 checks across 13 rules and 6 surfaces between 7 August and 25 October, which turned up 5,424 findings, and the rule that catches a follow-up echoing its own opener fires often enough to pay for itself.

After three touches the thread stays open and the clock gets long. A short question about timing goes out at 10 days and again at 45, capped at 250 characters, one question in it. Of 4,288 invitations in our log, 1,086 people accepted and 290 have replied so far, which leaves 896 people who said yes to the connection and have stayed quiet since. Those stay in play: 811 of them have had at least one message from us, and that pool is where the long clock earns its keep.

The volume a short sequence releases goes into the top. One profile sending three touches per person covers three times more people than one sending nine, at the same daily ceiling, and the opener is the touch paying 12.0%. Moving budget from touch six to touch one is the highest-return change available in a cadence.

And the measurement runs on cold touches only, forever. This is the part worth copying tomorrow whatever your numbers say: pull your sent messages, order them inside each thread, drop every message that followed an inbound reply, and count first replies against each position. Public ranges for the channel are collected on our cold outreach benchmarks page if you want something to read your own table against. The exercise takes an afternoon, and it answers the sequence-length question for your market with your own data, which beats any number in this article, ours included.

Questions buyers ask