learning-weekly-readoutlisted

# Weekly readout: one change, never on noise Once a week, slice replies by signal, segment and message variant. Decide nothing on a slice with fewer than 30 sends or fewer than 3 replies. Apply at most one change per week, so next week can tell you whether it worked. Judge a campaign by replies that left a next step, not by accepts. ## Why A good accept rate hides a dead bottom of the funnel, and the better the accept rate, the better it hides it. In our own data, one client's senders had the best accept rate in our roster: 361 invitations, 19.1% accepted, 16 replies, zero qualified leads. Another sender had a lower accept rate: 141 invitations, 15.6% accepted, 3 replies, 2 qualified leads. In the first campaign 12 of the 16 replies were "we already have this covered" or "not my area", and an offer to talk appeared in 38% of threads against 67% in the second. Across our whole engine at one point, 1465 touches in a month produced 32 replies (2.2%). The number that matters moves slowly, so changing two things at once leaves you with no idea which one moved it. ## How to apply 1. Tag every batch at creation with signal, segment and date, and every message with its variant. Without tags there is nothing to slice. 2. Pull the week: sends, accepts, replies, positive replies (replies that left a next step), meetings. 3. Compute reply rate and positive rate by signal, segment, message variant and lead source. 4. Use cohorts for decisions: count replies only within a fixed window