outreach

name

outreach

description

Run 1:1 outreach campaigns end-to-end — sourcing the list, enriching it, triaging into segments, setting up a Notion tracker, drafting templates, sending, and logging responses. Use this skill whenever Shaw is building a list of people to reach out to, setting up outreach tracking, or crafting the messages themselves. Triggers include: "find people on LinkedIn", "pull my connections", "build a lead list", "enrich this CSV", "look up where these people work", "get company sizes", "set up outreach", "create an outreach table", "add people to the outreach list", "who have I reached out to", "update outreach", "write me an outreach message", "draft a DM template", or any time Shaw shares a batch of names he's reaching out to. Also casual phrasing like "check my LinkedIn for analytics people", "I texted these 5 people", "add them to the list", "what should I say to these people". Covers workshops, events, research calls, partnerships, and any 1:1 outreach initiative.

Outreach

This skill runs Shaw's outreach campaigns end-to-end. Two halves of the same workflow: build the list, then work the list. A campaign is a coherent outreach push with a clear audience and goal — a workshop, an event, a research-call batch, a partnership push.

General principle: confirm before creating

Before committing to anything structural — a pipeline plan, a tracker schema, a segment scheme, a template direction — use the AskUserQuestion tool to present the plan and get Shaw's sign-off. Assumptions on shape are easy to get wrong and tedious to undo. A quick confirmation saves rework.

Pipeline design principles

Every campaign is a sequence of (enrich → filter) cycles using progressively more expensive tools on a shrinking candidate set. For any piece of information you need about a candidate, try these in order — only falling back to the next when the previous can't resolve:

1. Extract from data you already have (free)
2. Web search with Haiku agents in parallel (cheap)
3. Chrome profile lookup (expensive)
4. Ask the user (most expensive)

This pattern is fractal:

• Pipeline level: the entire campaign is enrich → filter cycles, cheapest first
• Stage level: within "get company names," first extract from titles (free), then web search the remainder (cheap), then Chrome the unknowns (expensive)
• Decision level: within "should we include this person," first check automated criteria, then enriched data, then ask Shaw

Kickoff alignment (do this FIRST, before any research)

The most important thing to lock at the start of every campaign — get explicit sign-off on all of it:

1. The wide net. What's the starting pool — who do we cast for and how (which source, which search terms)? This is the top of the funnel; agree on it before scraping a single name.
2. The non-negotiables. In most campaigns these are role and company size — the two axes that actually gate. Everything else (industry, AI-interest, location) is scored/flexible, not a hard cut. State the role bar (must own/lead or be budget-authority C-suite — decide explicitly whether hired C-suite counts, or founders/owners only) and the size band (e.g. 10–1,000 employees).
3. The anti-ICP. Ask "who looks like a fit but isn't actually a buyer?" and write it down. This is easy to miss and expensive to discover late — last time the realization that "AI-native builders/teachers are peers, not customers" surfaced mid-run and inverted the entire ranking. Decide exclusions up front.
4. The weights. If scoring, set component weights now (role + size dominate; flexible factors weigh less). Don't let weighting emerge through iteration.
5. Seed disqualified.md from the CRM — first. Pull Clients (Nurture) and Active Leads from the Notion CRM and exclude anyone already a client or in the active pipeline before scoring or enriching them. Don't leave CRM dedup to the end.
6. Sanity-check the channel. Is the warm network even the right well? An audience-saturated network (e.g. an AI educator's connections are ~half AI builders) will surface mostly anti-ICP under role-based search. Sourcing is often the binding constraint, not scoring — consider adopter/industry-oriented terms with the peer-exclusion baked in, or a different source entirely.

Then classify each criterion by evaluation cost (free / cheap / expensive / manual), sequence cheapest first, and present the pipeline to Shaw before starting.

Filtering discipline (when to cut vs. keep)

• Cut early only on cheap + unambiguous signals. If role or size is unmistakable from the headline alone — an obvious solopreneur/freelancer/one-person shop, an obvious employee at a mega-corp, a clear non-buyer (IC, student, ex/aspiring founder) — drop them at the free stage; there's no value enriching them.
• Keep everything ambiguous or unknown → enrich → filter on facts. A fuzzy headline, a title that could be founder-or-staff, or a missing value is NOT grounds for cutting. Carry them through enrichment and disqualify only once you have the data.
• Never filter on a field until it's fully populated. A missing company size is not "out of band" — it's unknown. Resolve it for everyone (web → Chrome) before applying any size cut. Err toward keeping the unknown until disqualified by real data.

Scoring & gating (for larger lists)

When a list is big enough to rank rather than hand-pick:

• Score, don't prematurely disqualify. Give each prospect a score from weighted components (e.g. role + size + fit). Soft criteria move the score; they don't eliminate.
• Gate the expensive step on the cheap score. Run the costly enrichment/inspection only on candidates already scoring well on the free/cheap dimensions — so you never spend Chrome time or per-profile lookups on someone who's clearly off-avatar.
• Set weights at requirements time (see Kickoff alignment). Role + size are the non-negotiable axes; flexible factors are scored but never the sole gate.

Enrichment efficiency

• Deduplicate enrichment targets. If 5 people work at Toyota, look up Toyota's size once.
• Extract before searching. Titles often contain companies ("PM at Figma"). Parse these before any web search.
• Batch web searches into parallel Haiku agents (~10 per agent), or a Workflow with structured output (agents read the lead file by index range and return validated rows — cleaner aggregation).
• Chrome is last resort — sequential and rate-limited. Reserve for the smallest possible set of unresolved unknowns.
• Finish the field before filtering on it. When a filter depends on company size, run web search then a Chrome cleanup pass on every remaining "unknown" — get to zero unknowns before any size cut. (Don't shortcut Chrome with a second web pass; last run that left 46 unresolved.)
• Validate coherence: size must be for the company they own/lead, not an employer or client. Watch for (a) pure employees — "GTM @ Stripe" means Stripe is not their company; (b) fractional/advisory/consultants — their led company is their own (often solo) practice, not the client they serve; (c) inflated or garbled web-search sizes. Verify role↔company↔size all refer to the same led entity.

Manual checkpoints

Place checkpoints after automated filters have narrowed the list but before expensive enrichment. Don't ask Shaw to review raw/unfiltered lists, and don't wait until the end for feedback — by then enrichment on cut candidates is already wasted. Shaw adds soft exclusions that no automation can know.

Disqualification tracking (disqualified.md)

A living document created at campaign start and updated throughout the run. Its primary job is intra-campaign efficiency: when Shaw gives feedback mid-pipeline ("[Person]'s off the table"), it goes into disqualified.md immediately and applies to all remaining candidates — preventing wasted enrichment on people already cut.

Two sections:

1. Disqualification principles — general rules applied automatically at every stage:

• "Exclude anyone already a client (Clients/Nurture) or in the active pipeline (Active Leads)" — seed this from the CRM at kickoff, before scoring or enriching
• "Exclude the anti-ICP" — the people who look like a fit but aren't buyers (decided at kickoff)
• "Only one person per company per job level"
• "No one obviously outside the size band (clear solopreneur, clear mega-corp employee)"

Checked before any enrichment step. If a principle disqualifies someone, skip the enrichment. (Soft/ambiguous size or fit is NOT disqualified here — that's scored and filtered on real data later.)

2. Disqualified individuals — people who pass all automated criteria but get cut for contextual reasons only Shaw knows:

Name	LinkedIn URL	Reason
[Person]	[linkedin-url]	Already reached out, not interested
[Person]	[linkedin-url]	Already have a contact at [Company]

Not for this table: People who fail criteria filters (wrong role, wrong size). Those should be caught by automation — if they slip through, fix the filter.

When Shaw gives feedback like "no, X is off the table because Y":

1. Add them to disqualified.md immediately
2. Check if the reason implies a new principle ("already have someone at [Company]" → "one contact per company")
3. If so, add the principle and retroactively apply it before continuing

---

The flow

1. Clarify the campaign and design the pipeline

Before touching any source, run the Kickoff alignment above (wide net · non-negotiables of role + size · anti-ICP · weights · CRM-seed disqualified.md · channel sanity-check). Plus:

• Campaign context: what's it for (workshop, event, research calls, partnerships)
• Target profile: role bar, seniority preference, location priority, size band (the hard one), industry (usually soft)
• Anti-ICP: who looks like a fit but isn't a buyer — decide and write down now
• Volume needed: how many final leads? Work backward assuming heavy attrition (a saturated network can attrit 95%+).
• Channels: LinkedIn DM, email, text, mix
• Segments: how will you group outreach (warmest first)? Common patterns: relationship proximity (in-person / virtual call / email-only), seniority × location, channel, fit tier.
• Exclusions: existing clients & active leads (from CRM), prior outreach, specific companies, geographic cuts

Then design the pipeline per the principles above — classify criteria by cost, sequence cheapest first. Present the plan to Shaw before executing.

Drop a reproducible recipe.md in the campaign folder capturing the agreed sequence (wide net → cheap filters → enrichment → hard filters → checkpoint → tracker → templates) so the run is repeatable and the next campaign can start from it.

2. Build the list

LinkedIn is the default source for cold/warm prospect lists → see linkedin-sourcing.md for the Chrome-based scraping workflow.

Other sources Shaw uses:

• CSV imports (past event attendees, conference lists)
• Contact form submissions
• Existing client/lead databases
• Batches Shaw shares directly in chat

Save raw leads to a working file so you don't keep 100+ people in context:

raw/<campaign>_leads_raw.json

3. Enrich (when the pipeline calls for it)

Enrichment runs between sourcing and triage when a filter depends on data you don't have yet. The most common case is company size — see lead-enrichment.md for the decision tree (1–9 contacts → Chrome direct; 10+ → parallel Haiku agents + Chrome cleanup).

Resolve the field fully before filtering on it. For company size: web-search pass, then Chrome cleanup on every "unknown" until none remain — then apply the size band. A missing value is not out-of-band; keep unknowns until real data disqualifies them. And validate coherence — the company you sized must be the one they own/lead, not an employer or client (see lead-enrichment.md).

Other enrichments follow the same pattern: extract → web search → Chrome → ask.

4. Triage into segments

Deduplicate by LinkedIn URL (or email, or whatever's unique). Then group by segment.

Segments are how Shaw prioritizes the list. The scheme varies by campaign:

• Relationship proximity: met in person → met virtually → have their email
• Seniority × location: DFW leaders → DFW managers → DFW ICs → non-DFW leaders → ...
• Channel: existing DMs → LinkedIn-only → cold email

Apply disqualified.md as part of triage. Present the triaged list to Shaw as a manual checkpoint before any further enrichment or tracker setup.

5. Set up the tracker

Once the list is confirmed, build the Notion tracker → see tracker-setup.md for the inline-database schema, segment legend, and page placement.

6. Draft templates

Each campaign needs templates — typically one per segment, sometimes per channel (text / email / LinkedIn DM). Start with the warmest segment first to establish the core pitch, then adapt outward. Draft one at a time so Shaw can react and steer before the next.

Save finalized templates to the campaign page in code blocks for easy copy-paste.

For writing craft specifically:

• Voice, tone, email mechanics (HTML, Gmail threading) → defer to email-helper
• Message craft — hooks, CTAs, personalization, reply friction → defer to conversion-copy (specifically channel-outreach.md)

This skill owns the operational side (who, when, which segment); message skills own the words.

7. Send, log, review

When Shaw adds contacts to the tracker, extract everything inferable from what he provides:

• Name (required)
• Segment (infer from context; Shaw usually specifies)
• Contact (email, phone, iMessage, WhatsApp, LinkedIn DM)
• Status (only if they've already hit a milestone)
• Last Contact (today if just reached out, else the date Shaw mentions)
• Notes (initial entry if there's activity)

Handle batches in a single notion-create-pages call.

When Shaw reports activity (replied, applied, booked call):

1. Find the contact in the tracker
2. Update Status and Last Contact
3. Append to Notes — never overwrite

Notes format: defer to the CRM skill — single source of truth. In short: short, comma-separated, date-stamped actions, always append.

Status review: fetch all entries and summarize as:

**Replies received:**
- [Name] — [brief context]

**Awaiting response:**
- [Name], [Name], [Name]...

**Next steps:**
- [Follow-ups due or next segments to activate]

Cross-reference Gmail when Shaw asks for updates — search each contact's email for new replies not yet logged, then update the tracker.

---

Relationship to other skills

• CRM skill — owns Notes format and pipeline-level lead management, including the promotion process. When an outreach contact graduates (books a call, asks about pricing), add them to the CRM and follow the CRM skill's "Promotion to main CRM" rule — the tracker entry becomes a historical breadcrumb (Added to CRM status, Next Contact cleared) while the main CRM owns ongoing touchpoints.
• email-helper — owns Shaw's voice, Gmail mechanics, reply/intro/decline templates.
• conversion-copy — owns message craft for conversion-focused copy. channel-outreach.md specifically covers cold-outreach DMs.
• notion-helper — adjacent; owns ad-hoc Notion creation outside structured campaigns.