/outbound-nurture-engine

name: outbound-nurture-engine
description: "Keep every outreach reply warm until it turns into a yes. Finds the leads your cold outreach did not convert yet ('not now', 'happy to connect', 'send me something', went silent after replying) and nurtures them on a 6-week cadence: picks the right case study, playbook or article per lead from your own content library (Notion, Google Drive, a website section or a pasted list), writes the picks onto the lead as custom attributes, then duplicates and fills your nurture campaign in La Growth Machine so touches, likes and follow-ups run natively. More positive replies, every campaign compounds, new content re-activates matching leads, content gaps say what to produce next. Use for: nurture my not-now leads, follow up on vague replies, re-engage leads that went quiet, build a nurture cadence, match content to leads, 'relancer mes leads pas maintenant'. Works from the LGM MCP or any outreach tool's CSV. For SDRs, founders, agencies, Heads of Sales. Maintained by La Growth Machine."
category: catch-opportunities
type: use-case
tags: [writing, analysis]

Outbound Nurture Engine

Turns the replies your outbound did not convert yet into a running nurture wave: each lead gets the content that fits them, written onto the lead in La Growth Machine, and a duplicated campaign carries the touches on a 6-week cadence.

Output discipline — read this first

When you run this skill, **return only the deliverables, nothing else.** No preamble, no narration of the steps, no restating these instructions. Each step is one framing line plus its table, code block or widget. If something essential is missing (where the content lives, which template campaign, which conversations), **ask one short, specific question and stop**. Nothing is written into La Growth Machine before the user has approved the wave.

Authority — read this first

• The **engine** (`scripts/build.py`) owns everything that can be silently wrong: discovering content URLs from a sitemap, keeping the library index, scoring content against leads, refusing to re-send a content, validating the custom-attribute sentences and the campaign messages. **Never do these by hand.** Run the engine; reason over its JSON. `python3 scripts/build.py --test` is the self-test.
• **Your job** is the judgment: reading each thread to label its situation and extract the lead's nurture profile, tagging the content you read, writing the per-lead sentences and the three campaign messages.
• `references/nurture-situations.md`: the four situations, the triage tree, what is *not* nurture. Read at Step 2.
• `references/library-schema.md`: the three content sources (Notion, website via sitemap, Google Drive), the tag axes, where the index lives. Read at Step 3.
• `references/touch-rules.md`: how to write the sentence stored on each lead and the three campaign messages. Read at Step 5.
• `examples/`: a fictional library and a fictional lead set for a worked run.

What it does

1. **Find** the leads worth nurturing in the replies you already got: not now, vague, ghosted after an exchange. 2. **Profile** each one from the full thread (pain, persona, industry, size, language, return condition). 3. **Index** your content library and **match** 1 to 3 contents per lead, never one already sent, with a generic fallback and a content-gap report. 4. **Write** one ready-to-insert sentence per content onto the lead (custom attributes) and add the lead to the wave's audience. 5. **Duplicate** the nurture template campaign, fill its three messages with the lead variables, assign the audience. The user launches it in La Growth Machine.

Out of scope, handed to sibling skills: a lead who spoke last and awaits an answer → `reply-draft-assistant`; a firm "not interested" or an unsubscribe → nothing; objection patterns across the inbox → `objection-analyzer`.

Workflow

Step 1 — Get the conversations

Three lanes. Whatever the lane, **read the full thread**, not the last message.

• **Inbox (LGM MCP).** `search_conversations` with `leadReplied: true`, optionally `campaignIds` or `lastMessageAtFrom` (default: last 90 days), `limit` 100, paginate with `searchAfter`. Returns ids and metadata only. Hydrate each kept conversation with `get_conversation_messages(conversationId)`. Skip `unsubscribed: true`. Capture `leadId`, `identityId`, `channel`, the lead's name from the thread, `last_received_at` and `last_sent_at`.

*If `search_conversations` rejects the call on argument types* (some clients pass every argument as a string and the tool wants arrays, booleans and numbers), take the campaign lane: `list_campaigns` for the campaign ids and the `identity.id`, then one `ask_your_outbound` query for the leads that replied:

  SELECT leadId, campaignId, MAX(date) AS last_reply FROM logs
  WHERE date >= DATE_SUB(CURRENT_DATE(), INTERVAL 120 DAY) AND campaignId IN ('<id1>','<id2>')
    AND type IN ('LINKEDIN_HAS_REPLY','GOOGLE_REPLY') AND leadId IS NOT NULL
  GROUP BY leadId, campaignId

then `get_lead_conversations(leadId, identityId)` and `get_conversation_messages` per lead. Same result, string arguments only.

• **CSV export** from any outreach tool: one row per conversation with lead id, name, last messages, dates. Ask for the last 3 to 6 months. No writes possible in this lane: the output is the matched contents and the sentences, for the user to paste as custom attributes.
• **Pasted thread(s)** for one or a few leads.

Bounded work: one search, one hydration pass, no re-reading.

Step 2 — Triage and profile (your judgment)

Apply `references/nurture-situations.md` to each thread. Label `situation` as `not_now`, `vague` or `in_nurture`; leave `ghosted` to the engine (it decides from dates). Extract the **nurture profile**: `pains[]` (the words the lead used), `persona` (function), `industry`, `company_size` (one of `1-10`, `11-50`, `51-200`, `201-1000`, `1000+`), `language`, `return_condition` (a date or an event