list-segmentation（列表分段）

Segment and Tier

Take an enriched table + hypothesis set and produce a tiered, segmented list. This decides WHO gets which message and in what order.

Inputs

Extruct API Operations

This skill delegates all Extruct API calls to the extruct-api skill.

For all Extruct API operations, read and follow the instructions in skills/extruct-api/SKILL.md.

The only Extruct operation in this skill is fetching enriched table data. Everything else is pure reasoning.

Workflow

Step 1: Load data

Use the extruct-api skill to fetch enriched table data. Parse all rows and their enrichment column values.

Read the hypothesis set file. Parse each hypothesis into:

• Number and short name
• Description with data points
• Best-fit company type

Step 2: Match companies to hypotheses

For each company row, evaluate which hypothesis fits best. Consider:

1. Enrichment data alignment — do the enrichment column values match the hypothesis's "best fit" description?
2. Signal strength — how many enrichment columns have useful data (not N/A)?
3. Specificity — does the company's profile match the hypothesis narrowly or broadly?

Assign each company ONE primary hypothesis. If multiple fit, pick the strongest signal.

Decision framework:

For each company:
  1. Read all enrichment values
  2. For each hypothesis:
     - Does the company's vertical/industry match the "best fit"?
     - Do enrichment values confirm the hypothesis pain point?
     - Is there a specific data point that makes this hypothesis resonate?
  3. Pick the hypothesis with the strongest evidence
  4. If no hypothesis fits well, mark as "Unmatched"

Step 3: Assign tiers

Three tiers based on fit strength and data richness:

Tier 1 signals (any of these):

• CEO/leadership made a public statement related to the hypothesis
• Recent news directly relevant to the pain point
• Hiring for roles that signal the hypothesis pain
• High hypothesis fit score from enrichment (grade 4-5)

Tier 3 signals (any of these):

• Most enrichment fields returned N/A
• No hypothesis match above threshold
• Company profile too generic to confidently segment

Step 4: Generate output

Output a segmented list in two formats:

Markdown table (for review):

## Segmented List: [Campaign Name]

### Tier 1 — [N] companies (personalized outreach)

| Company | Domain | Hypothesis | Tier Rationale | Hook Signal |
|---------|--------|-----------|----------------|-------------|
| [name] | [domain] | #[N] [name] | [why this tier] | [specific hook] |

### Tier 2 — [N] companies (templated outreach)

| Company | Domain | Hypothesis | Tier Rationale |
|---------|--------|-----------|----------------|
| [name] | [domain] | #[N] [name] | [why this tier] |

### Tier 3 — [N] companies (hold/re-enrich)

| Company | Domain | Issue |
|---------|--------|-------|
| [name] | [domain] | [what's missing] |

CSV (for email-generation):

Save to claude-code-gtm/csv/input/{campaign-slug}/segmented_list.csv with columns:

• company_name, domain, tier, hypothesis_number, hypothesis_name, tier_rationale, hook_signal

Step 5: Review with user

Present summary stats:

• Total companies: N
• Tier 1: N (X%)
• Tier 2: N (X%)
• Tier 3: N (X%)
• Unmatched: N

Ask:

• "Does the tier distribution look right? (Typical: 10-20% Tier 1, 50-60% Tier 2, 20-30% Tier 3)"
• "Any companies that should move tiers?"
• "Ready to proceed to email-generation?"

Reference

See references/tiering-framework.md for the detailed tiering decision matrix.