ironclaw-lead-enrichment（铁爪铅浓缩）

Lead Enrichment — Multi-Source Data Completion

Enrich CRM contact records by filling missing fields from multiple sources. Works with DuckDB workspace entries or standalone JSON data.

Sources (Priority Order)

1. LinkedIn (via linkedin-scraper skill) — name, title, company, education, connections
2. Web Search (via web_search tool) — email patterns, company info, social profiles
3. Company Website (via web_fetch) — team pages, about pages, contact info
4. Email Pattern Discovery — derive email from name + company domain

Enrichment Pipeline

Step 1: Assess What's Missing

-- Query the target object to find gaps
SELECT "Name", "Email", "LinkedIn URL", "Company", "Title", "Location"
FROM v_leads
WHERE "Email" IS NULL OR "LinkedIn URL" IS NULL OR "Title" IS NULL;

Step 2: Prioritize by Value

• High priority: Missing email (needed for outreach)
• Medium priority: Missing title/company (needed for personalization)
• Low priority: Missing education, connections count, about text

Step 3: Enrich Per Record

For each record with gaps:

If LinkedIn URL is known but other fields missing:

1. Use linkedin-scraper to visit profile
2. Extract: title, company, location, education, about
3. Update DuckDB record

If LinkedIn URL is missing:

1. Search LinkedIn: {name} {company} or {name} {title}
2. Verify match (name + company alignment)
3. Store LinkedIn URL, then scrape full profile

If Email is missing:

1. Find company domain (web search or LinkedIn company page)
2. Try common patterns:

- first@domain.com - first.last@domain.com - flast@domain.com - firstl@domain.com

1. Optionally verify with web search: "email" "{name}" site:{domain}
2. Check company team/about page for email format clues

If Company info is missing:

1. Web search: "{name}" "{title}" or check LinkedIn
2. Fetch company website for: industry, size, description, funding

Step 4: Update Records

-- Update via DuckDB pivot view
UPDATE v_leads SET
  "Email" = ?,
  "LinkedIn URL" = ?,
  "Title" = ?,
  "Company" = ?,
  "Location" = ?
WHERE id = ?;

Bulk Enrichment Mode

For enriching many records at once:

1. Query all incomplete records from DuckDB
2. Group by company (scrape company once, apply to all employees)
3. Process in batches of 10-20 records
4. Report progress after each batch:

   Enrichment Progress: 45/120 leads (38%)
   ├── Emails found: 32/45 (71%)
   ├── LinkedIn matched: 41/45 (91%)
   ├── Titles updated: 38/45 (84%)
   └── ETA: ~15 min remaining

1. Save checkpoint after each batch (in case of interruption)

Enrichment Quality Rules

• Confidence scoring: Mark each enriched field with confidence (high/medium/low)

- High: Direct match from LinkedIn profile or company website - Medium: Inferred from patterns (email format) or partial match - Low: Best guess from web search results

• Never overwrite existing data unless explicitly asked
• Flag conflicts: If enriched data contradicts existing data, flag for review
• Dedup check: Before inserting LinkedIn URL, check it's not already assigned to another contact

Email Pattern Discovery

Common corporate email formats by frequency:

1. first.last@domain.com (most common, ~45%)
2. first@domain.com (~20%)
3. flast@domain.com (~15%)
4. firstl@domain.com (~10%)
5. first_last@domain.com (~5%)
6. last.first@domain.com (~3%)
7. first.l@domain.com (~2%)

Strategy:

1. If you know one person's email at the company, derive the pattern
2. Search web for "@{domain}" email format
3. Check company team page source code for mailto: links
4. Use the most common pattern as fallback

Output

After enrichment, provide a summary:

Enrichment Complete: 120 leads processed
├── Emails: 94 found (78%), 26 still missing
├── LinkedIn: 108 matched (90%), 12 not found
├── Titles: 115 updated (96%)
├── Companies: 118 confirmed (98%)
├── Locations: 89 found (74%)
└── Avg confidence: High (82%), Medium (14%), Low (4%)

Top gaps remaining:
- 26 leads missing email (mostly small/stealth companies)
- 12 leads missing LinkedIn (common names, ambiguous matches)

DuckDB Field Mapping

Standard field names for Ironclaw CRM objects: