Skip to main content

​
Your First Campaign

This guide walks you through running your first outbound campaign using GTM Skills, from building your company context to sending emails. Expect 2-3 hours for a 200-company campaign.

​
What You’ll Build

By the end, you’ll have:
  • Company context file with ICP, voice, and proof points
  • 200+ qualified target companies
  • Enriched data for personalization
  • Segmented and tiered prospect list
  • 150+ generated emails ready to send

​
Prerequisites

1

Get Extruct API Token

Sign up at extruct.ai and grab your API token. Free trial includes 100 credits.
export EXTRUCT_API_TOKEN=your_token_here
2

Install GTM Skills

Via Claude Code plugin manager:
/plugin marketplace add extruct-ai/gtm-skills
/plugin install gtm-skills
Or via skills.sh CLI:
npx skills add extruct-ai/gtm-skills

​
Campaign Workflow

​
Step 1: Build Company Context

Start with the context-building skill to capture your ICP, voice, and proof points. 
I'm building [your-website.com].
Help me create a company context file.
Claude will ask for:
SectionWhat to provideExample
What We DoProduct one-liner, key numbers”API-first company search engine. 50M+ companies indexed.”
VoiceSender name, tone, constraints”From: Sarah Chen. Tone: technical peer. No superlatives.”
ICPTarget profiles with sizes”B2B SaaS, 20-200 people, US/EU, selling to enterprise”
Win Cases2-3 happy customers”Acme Corp uses us for supplier discovery. 10K searches/mo.”
Proof LibraryReady-to-paste PS sentences”PS. Shopify runs 50K lookalike searches/mo to find acquisition targets.”
The context file lives at claude-code-gtm/context/{company}_context.md and all other skills read from it.

​
Step 2: Build Your Target List

Use list-building to find 200-500 companies matching your ICP. Option A: Start from a win case (recommended) 
Find companies similar to acme.com (one of our best customers).
Target 200 companies, B2B SaaS, 20-200 employees, US only.
Option B: Semantic search 
Find 200 companies that:
- Build B2B marketplace platforms
- Serve enterprise buyers
- Need supplier/vendor discovery
Output: Extruct table with 200+ companies + auto-enriched company profiles.
Mix lookalike and semantic search for broader coverage. Run both in parallel, then deduplicate.

​
Step 3: Enrich for Personalization

Add 3-5 research columns to segment and personalize. 
Add these enrichment columns to table [table-id]:
1. Primary vertical (select: Manufacturing, Logistics, Healthcare, Fintech)
2. Platform type (text: what their product does)
3. Recent funding (select: Bootstrapped, Seed, Series A, Series B+)
Claude will:
  1. Create AI agent columns
  2. Run enrichment (1 credit per cell)
  3. Monitor progress
  4. Show you a quality spot-check
Typical enrichment time: 5-10 minutes for 200 companies × 3 columns.

​
Step 4: Generate Hypothesis Set

Create pain hypotheses based on your ICP and win cases. 
Read my context file and generate a hypothesis set
for marketplace platforms that need supplier discovery.
Output: claude-code-gtm/context/{vertical-slug}/hypothesis_set.md Example hypothesis: 
## Hypothesis 1: Stale supplier data

**Pain**: Marketplaces with 1000+ suppliers see 15-20% profile decay per year.
**Mechanism**: Companies rebrand, get acquired, go out of business.
**Best fit**: B2B marketplaces, 2+ years old, manufacturing/logistics vertical

​
Step 5: Segment and Tier

Match companies to hypotheses and assign tiers. 
Segment table [table-id] using the hypothesis set.
Tier by fit strength and data richness.
Claude produces:
TierCriteriaWhat happens
Tier 1Strong fit + data-rich + clear hookPersonalized email + review
Tier 2Medium fit or data-richTemplated email
Tier 3Weak fit or missing dataHold for re-enrichment
Typical distribution: 15% Tier 1, 60% Tier 2, 25% Tier 3.

​
Step 6: Find Decision Makers

For Tier 1 + Tier 2 companies, find the right people. 
Find decision makers at these companies:
- VP Sales, Head of Sales
- Revenue Operations, Sales Operations
- Marketplace Operations
Max 3 people per company.
Output: Extruct people table with names, roles, LinkedIn URLs. (2 credits per person found)

​
Step 7: Get Verified Emails

Enrich LinkedIn profiles with verified emails. 
Get verified emails for the people table using Prospeo.
Claude will:
  1. Ask for your Prospeo API key
  2. Run enrichment in batches
  3. Report hit rate (expect 60-80%)
  4. Save to claude-code-gtm/csv/input/{campaign}/contacts.csv
This step uses external API credits (Prospeo, Fullenrich, etc). Check pricing before running at scale.

​
Step 8: Build Email Prompt

Create a self-contained prompt template for this campaign. 
Build an email prompt for this campaign.
Target: VP Sales at B2B marketplaces
Angle: Stale supplier data (Hypothesis 1)
Claude reads:
  • Context file (voice, value prop, proof library)
  • Hypothesis set (pain mechanisms, evidence)
  • Enrichment columns (for personalization rules)
And produces: claude-code-gtm/prompts/{vertical-slug}/en_first_email.md

​
Step 9: Generate Emails

Run the prompt against your contacts CSV. 
Generate emails for contacts.csv using the prompt template.
For 150 contacts, expect 5-10 minutes. Output: claude-code-gtm/csv/output/{campaign}/emails.csv Quality checks:
  • Word limits respected
  • Enrichment data actually used (not generic)
  • Proof points vary per recipient
  • Example queries are specific to each vertical

​
Step 10: Review Tier 1 Emails

For Tier 1 companies, run response simulation. 
Simulate prospect review for all Tier 1 emails.
Claude generates a persona (CEO, VP Sales, etc) and reviews each email, flagging:
  • Generic phrasing
  • Weak hooks
  • Confusing value prop
  • Proof points that don’t resonate
You can refine and regenerate based on feedback.

​
Step 11: Upload to Sequencer

Load finalized emails into your email sequencer. 
Upload to Instantly. Campaign name: marketplace-supplier-q1
Claude will:
  1. Map CSV fields to Instantly lead schema
  2. Create or find campaign
  3. Upload leads with dedup
  4. Show pre-send checklist
Pre-send checklist:
  • Email accounts connected and warmed
  • Sending schedule set (timezone, daily limits)
  • Custom variables rendering correctly
  • Unsubscribe link present
  • Reply handling enabled
Claude will NEVER auto-activate your campaign. You must manually review and start it in your sequencer.

​
File Structure After Campaign

claude-code-gtm/
├── context/
│   ├── your-company_context.md
│   └── marketplace-supplier/
│       ├── hypothesis_set.md
│       └── sourcing_research.md
├── prompts/
│   └── marketplace-supplier/
│       └── en_first_email.md
└── csv/
    ├── input/marketplace-supplier-q1/
    │   ├── people_linkedin.csv
    │   └── contacts.csv
    └── output/marketplace-supplier-q1/
        └── emails.csv

​
After Sending: Feedback Loop

After 1-2 weeks, import campaign results. 
Update my context file with campaign results.
Here's the export from Instantly [attach CSV].
Claude will:
  1. Add a row to Campaign History
  2. Update Active Hypotheses (promote winners, demote losers)
  3. Update Proof Library based on new learnings
This creates a flywheel: each campaign improves the next.

​
Next Steps

​
Troubleshooting

Low enrichment hit rate (less than 50%)
  • Tighten your ICP - too broad means low data coverage
  • Try different enrichment prompts
  • Run a second enrichment pass with different agent type
Low email hit rate (less than 60%)
  • Try a different contact enrichment provider
  • Run a waterfall (two providers)
  • Focus on larger companies (better data coverage)
Tier 3 too large (more than 40%)
  • Add more enrichment columns before segmentation
  • Relax tier thresholds
  • Re-run list building with tighter filters

Build docs developers (and LLMs) love

Get started for freeTalk to us