Field note
Sales loops that wait for a human
Since early July this lab has run two sales loops side by side. One sources accounts from a licensed people database and from public job posts. The other works connection requests, DMs, and posts: twenty unpersonalized connects a day to people already commenting on category threads, a four-step DM, a daily proof post, and two lead magnets a week. Both loops stop at a queue. A human sends every message.
Signals start cheap. Company size, geography, role, and hiring news are enough for a first pass. A full dossier only happens after the account clears a written fit, timing, and intent threshold. On the network side, a comment on a category post is the warm signal. A one-word reply to a filter CTA is a hand raise. A comment on a magnet means send the resource and wait.
Qualification is three numbers written down. Fit, timing, and intent each get a score. The message type follows the public surface. Teardown if the site is weak, audit if there is something specific to inspect, direct if the problem is already obvious. Connection lists get marked keep or skip before anyone is added. Magnets use disqualifying language so the wrong company size and role stay out.
Proof comes from work that already shipped. A post needs a live URL, a screenshot, or a commit. Client numbers stay out unless that client signed off. Friday proof reloads every link the morning it goes out. If a URL is down, that line is cut. Teardown and opinion posts follow the same receipt rule. No receipt, no draft.
Drafts follow a fixed week. Monday teardown, Wednesday build-in-public, Friday proof. Each slot gets three platform variants. Daily posts weave one proof point and end with a filter CTA: who this is for, what they are stuck on, and one word to reply. DMs stay four separate messages. First a greeting. Then one compliment tied to a checkable fact. Then one pain question that hangs. Then one observation a peer would nod at. They do not collapse into a single pitch. The next message waits for a reply unless a full kit was asked for.
Follow-up sits in the same queue. After a connect is accepted, only the next message is written. After a magnet comment, the resource goes out with no pitch, then the same four-step sequence later. When impressions on a magnet slow, the same asset is reposted. Replies get a label and a next action. The labels are interested, objection, not now, unsubscribe, or auto. Nothing auto-replies. Unsubscribe never gets sequenced again.
Nothing sends until a person approves it. The agent does not post, schedule, mail, or click connect. Remaining credits are checked before a batch so a run cannot blow a cap. Calendar swaps are proposed with one line of rationale. The calendar does not move until the operator says so.
Each week the loop is scored against the last run. Correctness, craft, safety, whether it actually shipped, and whether anything was learned. A pass needs every dimension at least three, and at least one dimension better than last week, or a new failure class written down. A failed pass blocks the closeout. Thin motions get trimmed. One-off blasts skip that check, so they stay out.
What gets counted is a reply, a meeting, or a close. The monthly bar is twenty qualified meetings. Accepts, impressions, and list size only matter as ratios into those three. If they do not turn into a reply, a meeting, or a close, they are vanity and they get cut.
The useful output is a queue a person can empty, not a campaign that fires itself.
