Warm network first

Needs and produces

You need a connected LinkedIn account, a saved Ideal Customer page, and a Tone of Voice page or another defensible writing reference. Choose one signal type for the worksheet: engagement with your own posts, profile viewers and new followers, or public engagement with a competitor or peer post. Connected-account material and third-party public research are different sources and should not be mixed in the same worksheet.

The play produces one signal worksheet, evidence of the post or comment that put each person there, an AI fit judgment with a supporting quote, a fit score, and a warm sequence limited to qualified rows. It does not produce outreach to the whole engager table, a second parallel motion to the same audience, or an assertion that engagement itself establishes product fit.

Capture one signal and qualify it

Architect starts with Context and the selected account. It records the signal on the row: which post, what the person said, or the observed engagement event. The worksheet has one row source. If competitor research is also useful, create its own worksheet, then read it separately. This keeps the signal, source boundary, and later qualification understandable in Workbooks.

Add a fit judgment that compares the row with the ideal customer profile and includes the supporting quote, then add a fit score. Exclude the author's team, existing customers, and people already in an open conversation through score conditions rather than hand edits. Contact data belongs after the fit score and should run only for rows that passed the qualification rule.

Run a bounded sample and read individual results in every score band. The leading metric is qualified warm rows with evidence, not the raw number of likes or comments. Capture and qualification may be runtime-priced and capped per cycle, so the run result, not a request, establishes its actual cost.

Draft, validate, and enroll the qualified slice

The outreach skill reads the worksheet fields that contain the genuine signal before drafting. The first LinkedIn message names that signal in one line, adds one relevant point, and asks a light question. The message must remain grounded in the row. A generic line about finding someone in a search is not personalization, and an invented fact is not a valid substitute for an empty evidence field.

Build a short sequence, no more than three steps for warm people. Preview rendered messages for several real rows, including a thin-evidence row, and correct unresolved variables or weak fallback copy. A LinkedIn connection request is separate from a DM. A DM step sends only after connection acceptance, and every step needs copy that is safe when a row value is empty.

Publishing, enrollment, starting delivery, and a delivered message are separate states. Publish the reviewed sequence, enroll only the frozen qualified slice, and read the result. Starting delivery happens only when the user expressly asks to start. One person uses one sender and one sequence. A reply, decline, or unsubscribe ends their run across channels.

1. 1

   Read the audience context

   Use linkedin-warm-signals to confirm the account, saved ideal customer profile, and one signal type. The operator checks the selected source and its window before capture.
2. 2

   Build evidence in a worksheet

   Use list-building to create a single-source signal worksheet with evidence, fit judgment, score, and a gated contact-data stage. Read a bounded sample in Workbooks.
3. 3

   Draft warm copy

   Use outreach to write from the captured evidence. Preview the sender, recipient rows, and final text, including safe fallbacks, before publishing.
4. 4

   Enroll only qualified rows

   Publish the sequence and enroll the qualified selection. Check enrolled and excluded counts. Do not confuse enrollment with a started or delivered message.
5. 5

   Read the live state

   When a start was explicitly requested, read the sequence state and its sender pacing. Report the state proved by the last result and the worksheet and sequence links.

Credits and leading metric

The play's paid work is capture and qualification, which are bounded per cycle. The catalog describes them as runtime-priced and hard-capped. Publishing is free. Enrollment and start happen only after a cap. Inspect the current estimate before a paid live run and record the charged credits from the finished result. Do not copy a standing credit number into the play.

Track qualified warm rows with an evidence-backed reason, then review reply quality after delivery has started. A raw engager count can grow while fit is poor, so it is not the outcome to optimize. Report the signal, window, rows captured, qualified count, examples of reasons, sender, daily cap, expected pace, and current state: built, enrolled, started, or delivered.

When it goes wrong

If the captured audience is mostly peers, recruiters, vendors, or existing customers, adjust the qualification rule and resample. Do not enroll the whole table to discover fit later. If the row lacks a usable signal, either create a grounded evidence column or use fallback copy that remains truthful on its own. Never invent the missing personal detail.

When the sequence is enrolled but silent, read whether it was started, whether the sender is connected, whether sending hours apply, and whether a cap has been reached. A reply should stop the lead's run. If that state is unclear, read the conversation and sequence before retrying any external action. For an empty or failed worksheet value, diagnose its column and run through Find out why something failed.

Do not route third-party research through a connected account to avoid provider cost. Connected accounts read material they own; third-party profile and competitor-post research stays public research. A capped capture cell or a delayed send identifies pacing, not a reason to run an unbounded retry.