2026 report
What 85,730 cold messages taught us
Email vs LinkedIn, which touch works, what prospects actually answer, and the myths that did not survive the data.
Charles Perret, devlo.ch Data: 85,730 messages and 4,812 replies, 250 campaigns, 13 client workspaces, April 2024 to March 2026.
Download the PDF (English)All figures are descriptive. No campaign in this dataset randomised one version against another, so "associated with" never means "causes". A rule is kept only when it points the same way in two separate samples.
1. At the same prospect, LinkedIn starts more conversations
2,914 prospects received both a LinkedIn message and a cold email.
3 in 4
Of the prospects who said yes, 3 out of 4 said it on LinkedIn. Only 1 in 3 said it by email (a few said yes on both channels).
But LinkedIn also collects more no's: 5.5% negative replies per message, against 1.6% for email. And around 4% of LinkedIn profiles belonged to people who had already left the company. Use both channels; they reach different people.
2. Most yeses are born in the conversation, not in the sequence
2 in 3
Almost 2 out of 3 replies classified as "interested" were not the first reply. The prospect first asked a question, redirected to a colleague or said "not now". Someone answered by hand, and the yes came later.
The sequence opens the door. The conversation is where objections get answered and the meeting gets booked. How fast your team answers a reply that is not a yes matters as much as the sequence itself.
3. Which touch produces what
Email, share of all interest collected by each touch (cumulative): touch 1, 35%; touch 2, 52%; touch 3, 82%; touch 4, 90%; touch 5, 96%.
- Touch 2 is the weakest email. It has the worst ratio of no's to yes's and twice as many unsubscribes as touch 1. It is the one to rewrite.
- On LinkedIn, stop after two messages. The ratio of no's to yes's goes from 5 to 12 after the second message.
- Restarting a used list does not work. A "new" campaign sent to contacts already reached got 0.07% interest, against 0.32% for a real first touch.
4. One call to action beat all the others
Asking who to talk to ("who handles this on your side?") was associated with 3 to 4.5 times more interest. It was the only call-to-action trait that held in both samples.
Traits that showed no measurable effect: asking for a time slot, asking for an opinion, a P.S., a link or video in the body, ending with a question.
5. Short beats long
- Subject lines of 1 to 4 words did about 2.6 times better than 5 words or more.
- Emails of 150 words or more: 0 interest out of 4,859 sends.
6. What prospects actually answer
| First reply | ||
|---|---|---|
| Already equipped | 24% | 16% |
| Flat no | 16% | 14% |
| Redirect to a colleague | 15% | 16% |
| Too early | 9% | 8% |
| Interest | 9% | 7% |
| Not concerned | 8% | 15% |
Price is almost never the objection: under 2% of the reasons given in negative replies. The incumbent supplier is, about one time in three. A message that does not say how you differ from what they already use is asking the wrong question.
7. Myths that did not survive the data
- "Send on Monday or Tuesday." No difference that holds across samples.
- "68% of replies are refusals." That came from keyword matching. Read properly: 43% negative, 14% interest, 2.3% unsubscribes.
- "Open rates tell you what works." Raw open events double-count re-opens (57% raw against 28.8% real on one workspace), and tracking pixels can go silent for months.
How we use this
devlo runs outbound for B2B companies in Switzerland and Europe. We use these findings to design the campaigns we run for our clients. If you want to know what they mean for your market, we run a free audit of your qualified lead potential at devlo.ch.
Method: Lemlist export of 4 March 2026, 13 client workspaces. Replies classified by a language model and checked by hand on 220 stratified replies (86.8% accuracy; "interest" precision and recall 0.96). Post-reply manual sends removed from denominators. Holdout: 3 workspaces, 22% of replies. Full protocol available on request.