Playbook warm intros skill

Playbook: Warm Intros

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: the campaign's whole premise is that the prospect is already one hop from the client, and you need both the list and the line. The strongest version, and the one to build first, is "you used to work at a company that is now our client's customer".

Do not use when: you want a second contact at an account you already email (playbook-name-to-other-prospects), people who engaged with a competitor's or customer's posts (playbook-linkedin-engagement), or companies similar to a seed set (playbook-lookalikes).

One-line output: warm_intro_line = "you came up through Coca-Cola, who we work with now", rendering as Noticed you came up through Coca-Cola, who we work with now.

⚠️ Status: sources probed, no line graded

This is the least proven playbook in the library, and it is written that way deliberately.

• Member 2's source probe passed — 20/20 rows carried a dated matched past experience, a profile URL and a city, and 15/20 had moved on.
• Member 5's source probe passed — the investor filter returns a real result count.
• No member has produced a graded warm_intro_line.

Everything below §3 is a build spec, not a measured path. Treat the coverage and quality claims as unverified until you run one.

1. A family, not one playbook

A warm-intro signal is any evidence that the prospect is already one hop from the client, so email 1 can lean on a real connection instead of a cold pitch.

Seven members share the copy shape warm_intro_line. They do not share a data source. So this is one routing table plus one build recipe per member.

#	Member	What it means	Status
1	same-company alumni	Everyone who ever worked at the case-study company and has since moved on	live, same build as 2
2	current-customer alumni	The prospect used to work at one of the client's current customers	live, top priority, build this first
3	engagement overlap	The prospect engages with the same content the client's world does	live, but routed to playbook-linkedin-engagement
4	same metro	The prospect sits in the same city as the client, a named rep, or an upcoming trip	live
5	shared investor or advisor	A common investor, board member or advisor links the two companies	live
—	mutual connection	A first-degree connection in common	deleted. Do not re-add it, do not prototype it
7	event attendee	Both attended, spoke at or sponsored the same event	live as a custom scrape every time. There is no reusable recipe and there will not be one
8	website visitor	The prospect's company already visited the client's site	live as a per-client toggle, not a standing integration

Why member 1 was reframed, and why it matters

Member 1 originally meant "overlapped in time with someone on the client's team". That list is always too small, and it never once got run.

Reframed, it means everyone who ever worked at the case-study company and has since moved on — which is a large, buildable list from the same source as member 2.

The general lesson: a warm-intro definition that requires a time overlap between two specific people produces a list too small to campaign on. Widen to "worked there at all, and left" and the same data source suddenly supports a real segment.

2. Output contract

Field	Type	Example	Null?
warm_intro_line	string	you came up through Coca-Cola, who we work with now	yes, ""
warm_intro_member	enum	current_customer_alumni	no
warm_intro_evidence	string, QA only	Coca-Cola, Brand Manager, 2018-2021	yes

Abstain value: "".

Keep the member field. Seven members share one copy shape, and without it you cannot tell later which angle actually produced replies — which is the only way this family gets narrowed to the two or three members worth keeping.

⛔ Never name a real third party in copy without permission

The line says "who we work with now", not "we work with Coca-Cola" — unless the client has explicitly approved naming that customer.

Naming a client's customer in cold email is the client's decision, not yours, and it is the fastest way to turn a clever campaign into an awkward phone call. Default to the unnamed form.

Note this is the same posture playbook-lookalikes takes with case-study companies, for the same reason.

3. Build member 2 first

The source is a people search with a past-company filter set to the client's customer or case-study companies.

What the probe confirmed on 20 rows:

• a dated matched past experience,
• a profile URL,
• a city,
• and 15 of 20 had moved on — which is the population you actually want.

That last number is the one to plan from. Roughly three quarters of "ever worked there" is "worked there and left", and the rest are current employees who must be excluded (they are the customer's staff, not prospects).

The exclusions this family needs

Inherit them from playbook-linkedin-engagement, because the shape is identical:

• Drop anyone currently employed by a source company. They work for the client's customer; they are not a warm intro, they are the customer.
• Never include the client's own domain in the source set. If it appears, the list is wrong — stop.
• Push the source companies onto the client's do-not-contact list, so this run's exclusions persist into future lanes.

4. Verification

Untested for the line. Sources probed only. See the status box above.

The first person to run a member owes the file: a real fill rate, a graded sample of at least 10 lines, and the member name on every row.

5. Clay implementation

• clay-table.md — the labelling half as columns.
• clay-workflow.md — the CLI-buildable version.

6. The line

The line is short, and the copy risk is entirely in what it names, not in how it is written.

Shape: you came up through <former employer>, who we work with now.

Rules:

• Lowercase first word, no trailing period, no em dashes, under 90 characters.
• Name the prospect's former employer (a public fact from their own profile) — not the client's customer relationship in identifying detail, unless approved.
• Never state a date. "In 2018 you were at X" is both creepy and frequently wrong when a profile is imprecise.
• Never imply you know someone in common unless you do. The deleted mutual-connection member exists as a warning: that claim is unverifiable at scale and reads as a lie when it is wrong.

Verification for this family is source-side, not model-side. The facts come from the prospect's own profile, and the model only phrases them — so the risk is a stale or mismatched profile, which no verifier over your own output can catch. Prefer sources that return a dated experience you can check.

7. Edge cases and failure modes

Symptom	Cause	Fix
The list is tiny and not worth a campaign	The member was defined as a time overlap between two specific people	Widen to "ever worked there, and has since moved on"
The list is full of the customer's current staff	No current-employment exclusion	Drop anyone currently employed by a source company
The client's own employees appear	The client's domain got into the source set	Stop. The list is wrong
The same people reappear next quarter	Row drops only fix this run	Push the source companies to the client's DNC list
A client is unhappy that their customer was named	The line named a real third party	Default to the unnamed form. Naming is the client's decision
The line states a date that is wrong	Profile date ranges are imprecise	Never state a date
Replies cannot be attributed to an angle	Seven members share one copy field	Keep warm_intro_member on every row
Someone proposes rebuilding mutual connections	It is a deleted member	It is unverifiable at scale and reads as a lie when wrong. Do not re-add it
An event member becomes a maintenance burden	It was turned into a standing recipe	Event overlap is a custom scrape every time. That is the decision, not a gap

Hard rules

• Build member 2 first. It is the only one with a passing source probe and a real population behind it.
• Never name a client's customer in copy without explicit approval.
• Never claim a mutual connection.
• Inherit the source-company exclusions, both halves.
• No member ships a line until at least 10 rendered lines have been read and graded.