Introduction
Most outbound sequences mix email and LinkedIn. So do most analyses. We analysed 25,416 messages to 14,551 people over four months (May to August 2026), separated by channel. The result is clearer than we expected.
The path matters, not the single message
Here we compare people, not messages. The question is: if I approach someone this way, how likely am I to get a reply?
| First-contact path | people | replies | reply rate |
|---|---|---|---|
| Email only | 6,820 | 170 | 2.5 % |
| Connection request sent | 7,237 | 550 | 7.6 % |
| of which accepted | 2,166 | 455 | 21.0 % |
| of which not accepted | 5,071 | 95 | 1.9 % |
| LinkedIn message, already connected | 494 | 72 | 14.6 % |
The connection request gets three times as many replies as email alone. That also holds for the replies we actually want: requests for a meeting or for information sit at 1.05 % against 0.45 %.
The second row matters as a whole. We count the connection requests nobody accepts. Showing only the accepted ones (21 %) flatters the number.
The connection request is the filter. Whoever accepts has already said yes to a conversation.
Whoever does not accept is an email contact
The least noticeable row is the fourth. People who do not accept the connection request reply at 1.9 % afterwards. That is worse than email alone.
For the sequence this means: after an unaccepted connection request, do not keep waiting on LinkedIn. Switch to email earlier.
A correction: the third contact
On 18 September we wrote that the third contact is the weakest step in the sequence. The numbers back then, across 37,375 messages: 5.06 % in step one, 2.10 % in step three. In the period of this analysis the mixed numbers look the same: 3.6 % in step one, 2.0 % from step three on.
That does not hold. In step three, 89 % of our messages are emails; in step one only 48 %. We measured the channel, not the step. Separated, it looks like this:
| Step | ||
|---|---|---|
| 1st message | 1.28 % | 5.81 % |
| 2nd message | 1.58 % | 9.95 % |
| 3rd message | 1.44 % | 6.09 % |
Within one channel the third step does not drop. The difference to the first two steps is not significant on either channel. This table counts single messages, which is why the rates sit below the people table above.
Statisticians call this Simpson's paradox. We call it: whoever analyses email and LinkedIn in one number ends up measuring the channel.
What counts within a channel
Two findings that only became visible after the split:
- No question in the connection request. Without a question, replies are about 45 % more frequent. In email a question makes no difference.
- No link in the first email. Without a link, replies are about 60 % more frequent. Our connection requests practically never carry a link, so the question does not arise there.
Both figures come from a sample of all replies against randomly drawn messages without a reply. The ratios hold; absolute reply rates cannot be read from them.
One caveat belongs here: we only send a connection request to people with a maintained LinkedIn profile. The two groups are not randomly assigned, and part of the lead comes from selection. In our view that does not explain a factor of three.
What we do with it
- We start with the connection request when there is a LinkedIn profile.
- We switch to email when the connection request is not accepted, instead of following up on LinkedIn.
- We write the connection request without a question and the first email without a link.
Next steps
We are measuring how long to wait for acceptance before switching to email. The raw data points to a few days. We will publish it once the period is large enough.
Goal
We want to choose the channel per person, not per campaign. Happy selling, Axel