I A/B tested 21 cold email first-line variants across 9,600 sends to VP-level buyers and two formats beat everything else by at least a full reply rate percentage point
firstline_testing_jas · 2026-09-10 · 0 views
First lines are where cold emails win or lose. Not the subject line, not the CTA. The first line. I've been running structured first-line tests for eight months across client campaigns targeting VP-level B2B buyers. Most of the popular advice in this community is wrong about what actually works at this seniority level.
I tested 21 variants across 9,600 sends over three months. Here's what the data showed.
The two formats that won consistently:
Format one: a specific operational observation that proves you understand their actual job. Something like: Most VPs of Sales I talk to at Series B companies are running forecasting manually in spreadsheets outside their CRM. That sentence asks no question, makes no claim about your product, and earns a pause from anyone who recognizes the problem. Reply rate on this format: 3.8 percent average across eight campaigns.
Format two: a direct reference to a public signal tied to their company specifically. Not a generic compliment about their recent funding round. Something grounded in what that event actually means operationally. Your team went from 12 to 31 in the last six months according to LinkedIn. That kind of scale usually creates specific infrastructure problems before it creates obvious ones. Reply rate: 3.6 percent average.
What lost consistently:
Any opener that starts with I. The word I in position one puts the reader in receiving mode before they've decided to engage. Cut it.
Openers referencing mutual connections who aren't real or recent. Buyers at VP level have been burned by fake mutual connections enough times that any name-drop they can't verify instantly moves to suspicion.
Questions as first lines for this ICP. Questions work below the director level. VP-level buyers respond better to observations than to questions. Save the question for line three.
I run all first-line testing through Smartlead with variant rotation enabled. Each test runs until I have at least 400 sends per variant before I draw conclusions. Under 400 per variant and the noise overwhelms the signal. I use 16 PuzzleInbox Google Workspace inboxes for these tests so infrastructure variation doesn't corrupt the results.