Short follow ups won 26.3 percent replies to 16.8, and the check row came back level

    Who this was forA B2B company with four people sending LinkedIn outreach, six weeks into the channel.

    An A/B result we could trust, and the report that carries it

    All case studies

    The drawing

    About 25 campaigns,four sendersSix A/B pairsCheck row: acceptanceReplies and positiverepliesSecond check: everycampaign split bylengthSix page channelreport

    The problem

    Six A/B pairs were running. Both sides of each pair sent the same connection note. The test was the first message after someone accepted: a pitch of 63 to 79 words against a single question of 12 to 15 words.

    LinkedIn A/B numbers are easy to over read. Pairs share lists. Arms end up different sizes. Counts get small fast.

    What we built

    A channel report for the client's leadership, six pages, with one rule for every number in it. Acceptance comes from real accept events, not network status. Reply rate divides by conversations started, not messages sent. Any small sample result is labeled as one and called not actionable.

    For the A/B section we added a check row. The connection note was the same on both sides, so acceptance should come back level. If it didn't, the test was measuring something other than the copy.

    What it produced

    MetricLongShort
    Acceptance, the check27.7%29.4%
    Reply rate16.8%26.3%
    Positive reply rate0.9%5.1%
    • The check row came back level, so the two rows under it hold.
    • A second test ignored the pairs and split every campaign by message length. Short still won: 25.5 percent replies to 17.3, and 5.0 percent positive to 1.1.
    • The number of people replying no was the same on both sides. Short copy wasn't just earning faster refusals.
    • Targeting showed up too. Head of titles accepted at 31.6 percent, Directors at 21.5.
    • The report's structure became the template for the client's future channel reports.

    Your scenario

    If you run A/B pairs in HeyReach or lemlist, add a row you expect to come back level before you read the rest. Our Sequencer Event Poller on Apify pulls accepts and replies into one table, so you can split them any way you like. Send us your pair results and we'll tell you whether we'd trust them.

    Related tools: HeyReach, lemlist, Apify, Sequencer Event Poller on the Apify Store, Sequencer Event Poller

    Some of the links on this page are affiliate links.

    Client names are always kept confidential at MMG, so they have been removed from this case study.

    Published August 25, 2026

    Last updated September 14, 2026