Playbook: LinkedIn Engagement skill

Playbook: LinkedIn Engagement

All rules here are best practice, not law. Override any of them when the campaign calls for it; note the best practice once and move on.

Use when an account's audience is the client's audience — a competitor or the client's own customer.

Do not use for more companies in a market, similar companies (playbook-lookalikes), or job changers (playbook-new-in-role). The chain is the same for both lanes; only the target URL changes.

Output: one row per engager — who, employer, post, evidence URL. engagement_line is opt-in.

The number that governs planning

⚠️ Roughly 1 in 10 raw engagers survives. Harvest 10 to 15x what you need.

Cost follows from that: about $4.45 per 1,000 raw engagers, which is about $45 per 1,000 usable prospects at the measured survival rate. That is squarely in expensive territory, so shortlist posts by engagement count before you scrape.

The source-company rule (both halves, always automatic)

This is the rule that separates a usable engagement list from an embarrassing one, and it has two halves that people implement only one of.

(a) Drop every engager employed by ANY source company — not just the author of the post they engaged with. Match each current employer on resolved domain first, then profile URL, then squashed name, and also on resolved email domain.

(b) Push every source company onto that client's do-not-contact list. Row drops fix only this run; the block list runs at send time, which is what stops the same people arriving through a different lane next month.

It bites hardest on the customer lane, where the source companies are people the client already works with.

⛔ Never block the client's own domain. If it appears in the source set, the list is wrong — stop.

Before any paid call

1. The operator confirms the source list. Say it out loud: "these companies and everyone who works at them go on this client's DNC list."
2. Resolve every source account to a bare domain first. An unresolved source silently disables both halves of the rule above. Unresolved means stop, not continue. (playbook-social-link-finding does this conversion in both directions.)
3. If the client's ICP is unknown, ask. Never infer a headcount band, a country list, or a title set.

Run it

0   Confirm source URLs, client ICP, and whether the brief wants a line (default: no).
0b  Resolve every source account to a bare domain -> source-companies.json (EXACT match only).
    Unresolved => STOP.
1   Company posts, windowed to the last month, shortlisted by engagement count.
2   Post reactions + post comments for the shortlisted posts.
3   ICP enrichment (headcount, company) on the engagers.
4   Deterministic pre-gate: source-company drop, ICP filter, dedupe.
5   Suppression pass: client DNC, existing TAM, live campaigns.
6   DNC push (once per harvest).
7   Email finding, valid only.

Rows are the deliverable; lines are opt-in. Never run a line writer unless the brief asks. When it does, an empty-line row leaves that campaign for the normal one rather than shipping a gap.

Verification

PASS 17/18 correct outcomes on the source-and-gate test — and note what that counts: a correct exclusion is scored as a usable outcome. Raw signal yield was 1 usable prospect in 17 unique people.

Both numbers are true and they measure different things. The gate works; the yield is thin. Plan from the yield.

⚠️ The end-to-end harvest has never been run in one pass. Do a one-post run and read the output before scaling.

Clay implementation

• clay-table.md — where the gate lives in a table.
• clay-workflow.md — the CLI-buildable version.

Edge cases and failure modes

Symptom	Cause	Fix
The list contains the client's own employees	The client's domain got into the source set	Stop. The list is wrong
The list contains the competitor's own staff	Only half (a) was implemented, matched on name alone	Match on resolved domain first, then profile URL, then squashed name, then email domain
The same people reappear next month through another lane	Half (b) was skipped	The DNC push is what persists. Row drops do not
The source-company rule appears to do nothing	An unresolved source URL produced an empty source set	An empty set silently disables both halves. Unresolved means stop
Cost per usable prospect is 10x the estimate	You costed raw engagers, not survivors	~1 in 10 survives. Harvest 10-15x
A person's employer is wrong	Taking the first entry of an experience array rather than the current one	Read the current experience explicitly
Rows silently vanish on a join	Post and engager records key differently	Confirm the join key on one post before scaling
The run stalls waiting on a scraper	There is no fallback source by design	The run waits. A different scraper is not a drop-in substitute here

Hard rules

• Both halves of the source-company rule, every run, automatically.
• The operator confirms the source list before any paid call.
• Rows are the deliverable. Lines are opt-in.
• If the ICP is unknown, ask.