Teardown: We Sent 100 Connection Notes Five Different Ways. Here's What Booked Meetings.

A Banyan customer ran an experiment over three weeks. Same ICP, Heads of Sales at Series A SaaS in India, ~3,500 prospects in their list. They split the list five ways and ran a different outreach motion on each cohort of 100. We tracked accept rate, reply rate, and meetings booked from each.

The numbers below are unedited. Names are anonymised.

Cohort A, Cold template (the control)

Standard cold connection note. No signal, no personalisation beyond first name and company.

“Hi {firstName}, came across your profile, you’re doing interesting work at {company}. We help SaaS sales leaders book more meetings with warm LinkedIn outreach. Open to connecting?”

Result:

• Accept rate: 28%
• Reply rate (of accepts): 4%
• Meetings booked: 1

The control. Roughly in line with what every cold-template campaign produces in 2026.

Cohort B, Cold template + AI personalisation

Same intent, but each note had an AI-generated first line referencing the prospect’s recent post or company news.

“Hi {firstName}, saw {company} just announced your Series B, congrats. We help SaaS sales leaders book more meetings…”

Result:

• Accept rate: 42%
• Reply rate (of accepts): 6%
• Meetings booked: 3

Better. But the second sentence was still a templated pitch, and prospects clocked it. Multiple replies were variants of “this felt automated.”

“

The personalised opener gets you in the door. The templated middle slams it shut.

Cohort C, Warm signal (post engagers), cold copy

People who had reacted to or commented on the customer’s posts in the last 30 days. Same templated body as Cohort A.

Result:

• Accept rate: 71%
• Reply rate (of accepts): 8%
• Meetings booked: 5

The accept rate jumped huge, they recognised the customer because they’d seen the post. But the templated body still capped reply rate. The lesson: warm signals fix the connection step but don’t fix the conversation.

Cohort D, Warm signal + signal-referencing copy

Same warm pool as C, but every note referenced the specific post they engaged with.

“Hi {firstName}, saw you reacted to my post about SDR comp last week. The thread got into whether base+commission still works for India SaaS, what’s your take? (We’re rebuilding ours and I’d love to compare notes.)”

Result:

• Accept rate: 73%
• Reply rate (of accepts): 24%
• Meetings booked: 14

The breakthrough. Same accept rate as C, but the reply rate tripled because the conversation actually started. Most replies were “open to talk” or pure substantive engagement (their actual take on the question).

Cohort E, Warm signal + signal copy + sequenced follow-up

Same as D, with a thoughtful follow-up 3 days after connection (referencing a different post or piece of content) and a second follow-up 8 days later.

Result:

• Accept rate: 73%
• Reply rate (of accepts): 31%
• Meetings booked: 18

The follow-ups added 4 meetings on top of D’s 14, modest extra lift, real but smaller than the jump from C to D.

The numbers, side by side

Cohort	Treatment	Accept	Reply (of accepts)	Meetings	Reply per 100 sent
A	Cold template	28%	4%	1	1.1
B	Cold + AI opener	42%	6%	3	2.5
C	Warm signal, cold copy	71%	8%	5	5.7
D	Warm signal + signal copy	73%	24%	14	17.5
E	D + thoughtful follow-up	73%	31%	18	22.6

What we expected vs. what we found

We expected the AI opener (B) to be the big jump. It wasn’t, the small lift over A was barely worth the effort.

We expected warm signals (C) to be the unlock. They drove accepts up 2.5x, but reply rate barely moved.

The actual unlock was the combination, warm signals plus signal-referencing copy in the body of the message (D). That’s where the reply rate tripled. From there, follow-ups (E) added a real but smaller increment.

The implication for anyone running warm outreach: don’t half-do it. Capturing warm signals and then sending templated copy gives you 5.7 replies per 100. Capturing warm signals and writing in your voice referencing those signals gives you 17.5. The pipeline difference is 3x for the same volume.

What we’d run differently

• Cohort F (next test): warm signal + signal copy + a human-written second message (instead of templated follow-up). Hypothesis: pushes E to ~25 meetings per 100.
• Cohort G: warm signal but with a deliberately weaker personalisation (mention the post topic but not the specific content). Hypothesis: closer to C than to D, confirms that specificity is the load-bearing part.

If you run either, send us the numbers. We’ll publish what comes back.