Prospecting — B2B Lead Generation from Existing Customers

Overview

Turn existing customers into a search template → find similar businesses on Google Maps → enrich → score → output actionable call lists.

One line: Known customer → profile → Maps search → enrich & rank → CSV call list + JSON index

Compliance & Legal Boundaries

Before running this skill, confirm:

Item	Requirement
Geography	Only collect businesses in jurisdictions where B2B outreach is lawful
Data source terms	Comply with Google Maps Terms of Service, Yelp ToS, and any site scraped
Rate limiting	Add reasonable delays between requests; do not hammer the same site
Do-not-call / anti-spam	Respect opt-out lists and local telemarketing rules
Privacy laws	GDPR / CCPA / PIPL: avoid collecting unnecessary personal contact info
Data accuracy	Only use publicly listed business contact info; do not enrich personal mobile numbers or home addresses

This skill collects publicly visible business information only. Do not use it to harvest personal emails, mobile phones, home addresses, or non-business contacts.

Data Retention & Storage

Rule	Implementation
Output path	prospect-data/{batch}/ under the user's workspace
Retention	Delete or archive prospect files when the campaign ends or they are no longer active
Sensitive data	Do not store personal mobile numbers or non-business contacts
Sharing	Do not share raw JSON/CSV with third parties without consent
Encryption	Move completed batches to encrypted storage if required by your policy
Access	Treat call lists and profile JSON as internal sales data; restrict access to authorized users

When to Use

• User gives a customer name + location and asks to find similar businesses
• User asks to build a prospect/call list
• User wants to find new clients in a specific industry (auto body, manufacturing, HVAC, etc.)

Input Required

Field	Required	Notes
Company name	✅	Core search term
Location (city/state)	✅	Search center point
Product purchased	❌	Helps with profiling

Even minimal input ("Bob's Auto Body, Orange CA") can start the full flow.

Execution Flow

Step 0: Self-Diagnostic & Adaptive Search Design (NEW — Auto-Optimization)

Before executing searches, the skill automatically analyzes search parameters and adjusts strategy without user intervention:

0.1 Coverage Gap Detection

After initial search round completes, automatically check for these warning signals:

Warning Signal	Threshold	Auto-Action
Zero results for keyword+center	0 listings returned	Swap keyword (e.g., "paint shop" → "auto paint" / "car paint" / "collision")
Low unique yield per center	<5 unique businesses per center after dedup	Expand radius or add satellite center
High non-target ratio	>30% listings are wrong industry	Tighten keyword (e.g., "paint shop" → "auto body paint" / "collision paint")
Chain under-representation	0 chain brands in results	Add brand keywords (Caliber, CARSTAR, Maaco, Gerber)
Equipment-related missing	No spray booth / frame machine in results	Add equipment keywords to second pass

0.2 Keyword Auto-Adjustment Rules

IF center + "paint shop" returns <5 valid auto-body listings:
  → REPLACE with "auto paint shop" OR "car paint" OR "collision paint"
  
IF center + "auto body shop" returns 0 results:
  → TRY "body shop" OR "collision repair" OR "auto repair"
  
IF chain brands known in industry but missing from results:
  → ADD "[brand] + city" as explicit search (e.g., "Caliber Houston")

0.3 Multi-Pass Search Protocol

PASS 1: Core keywords (auto body shop, collision repair, paint shop)
  ↓
Auto-analyze coverage gaps
  ↓
PASS 2: Gap-fill keywords (adjusted based on PASS 1 results)
  ↓
PASS 3: Equipment/brand keywords (spray booth, frame machine, Caliber, CARSTAR)
  ↓
Final dedup + scoring

No user input required — the skill self-diagnoses and adjusts between passes.

Step 1: Profile the Existing Customer (8-step fixed process)

Read references/profiling.md for the full 8-step process. Key actions:

1. Google Maps deep dive — Use agent-browser to search [company name] [location], extract: address, phone, rating, review count, business type, hours, website, photos, chain status
2. Review sampling — Sample reviews with keyword filtering (not all reviews). Generic keywords: new, expand, equipment, upgrade, install, moved, bigger + industry-specific keywords (e.g., for auto body: paint booth, insurance, fleet, dealer)
3. Social/web enrichment — Only for 🔴 chain (FB+LinkedIn+website) or 🟡 mid-tier (FB+website). Skip 🟢 small (no website)
4. Output a Profile Card — Standard format saved to prospect-data/{batch}/profile-{name}.json

Tier detection (determines enrichment depth):

• 🔴 Chain/large: name contains chain markers OR >200 reviews
• 🟡 Mid-tier: has website, 50-200 reviews
• 🟢 Small: no website, <50 reviews

Step 2: Maps Batch Search (agent-browser automated, with self-optimization)

Read references/search-strategy.md for the complete search framework.

NEW: Self-Optimizing Search Loop

The search now runs in 3 automatic passes with gap detection between each:

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   PASS 1        │     │   PASS 2        │     │   PASS 3        │
│   Core keywords │────→│   Gap-fill      │────→│   Equipment/    │
│   (4-6 per      │     │   (auto-adjusted│     │   Brand deep    │
│    center)      │     │   based on P1)  │     │   dive          │
└─────────────────┘     └─────────────────┘     └─────────────────┘
       ↓                       ↓                       ↓
   [Auto-analyze]          [Auto-analyze]           [Final dedup]
   Coverage gaps            Remaining gaps           + Score

Pass 1: Core Search (always runs)

• Multi-center: Large cities (>2M) use 4-6 search centers
• Keyword matrix: 4-6 keywords per center (core + service + equipment + brand + scene)
• Pagination: Scroll and load 3 times per search
• Deduplication: Cross-center, cross-keyword deduplication

Pass 2: Gap-Fill (auto-triggered if gaps detected)

After Pass 1 completes, automatically check:

Check	Condition	Auto-Action
Zero-result centers	Any center returned 0 listings for >2 keywords	Add 1-2 satellite centers (midpoint between existing centers)
Low-yield keywords	Any keyword returned <3 valid listings across all centers	Replace with synonym (see Keyword Swap Table below)
Missing chains	Known chain brands not found	Add explicit brand+city searches
Geographic holes	Large gaps between centers	Add midpoint center

Keyword Swap Table (auto-applied):

Original (low yield)	Replacement 1	Replacement 2	Replacement 3
paint shop	auto paint shop	car paint	collision paint
auto body shop	body shop	collision repair	auto repair
collision center	collision repair	auto body repair	crash repair
frame machine	auto frame	chassis repair	structural repair
spray booth	paint booth	auto paint booth	HVLP booth

Pass 3: Equipment & Brand Deep Dive (auto-triggered)

If target industry involves equipment (auto body, manufacturing, HVAC), automatically add:

• Equipment keywords: spray booth, frame machine, car lift, CNC mill, etc.
• Brand searches: [Brand] + [City] for known chains (Caliber, CARSTAR, Maaco, Gerber, Crash Champions)

Search execution:

1. For each center point × each keyword: open Google Maps, extract listings, paginate 3x
2. Collect: name, phone, address, rating, review count, business type, website status, chain markers
3. Dedup: same name + same address = duplicate
4. Remove: permanently closed, non-target industry

Save to: prospect-data/{batch}/candidates-raw.txt (raw extraction log) + candidates.json (deduplicated)

Auto-optimization logging: After each pass, append to candidates-raw.txt:

[PASS-N-ANALYSIS] Center: X, Keyword: Y, Results: Z, Action: [none|swapped|expanded|added-brand]

Step 3: Auto-Tier Candidates (with self-validation)

Based on Maps data, assign tiers. Chain stores are NOT excluded — they are valid prospects with a different approach strategy.

NEW: Self-Validation Checklist (auto-executed before tiering)

Before finalizing candidates, automatically verify:

Check	Action if Failed
Duplicate names with different addresses	Mark as chain locations, keep all
Same name + same address appearing multiple times	Deduplicate, keep most complete
Business type mismatch (e.g., "car wash" labeled as "collision")	Re-classify or flag for manual review
Phone placeholder pattern (555, 000, 1234)	Flag as phone_status: unverified_placeholder
Address in different city than search center	Verify: satellite location vs data error

Auto-chain detection (no manual list needed):

IF name contains: Caliber, CARSTAR, Maaco, Gerber, Crash Champions, ProColor, 1st Choice
  → Mark as chain_brand: [detected brand]
  
IF same name appears in >2 different addresses across centers
  → Mark as chain_brand: [normalized name]
  → Set tier: "连锁-XX"

Tier	Criteria	Next action
🔴 Chain/large	Chain name OR >200 reviews OR multi-location detected	Deep enrichment + chain procurement strategy
🟡 Mid-tier	Has website, 50-200 reviews	Medium enrichment
🟢 Small	No website, <50 reviews	Skip enrichment

Chain store prospecting strategy — Read references/chain-strategy.md for the full three-call approach:

• Call 1: Local store — NOT to sell, but to identify procurement decision chain
• Call 2: Regional/corporate — pitch to the person who can approve multi-location deals
• Call 3: Follow-up with proposal

Key principles:

• Chain stores have large, stable equipment needs — one deal can cover multiple locations
• Local store manager is the entry point, not the decision-maker (usually)
• Key question: "Is equipment purchasing handled locally, or should I speak with your regional/corporate procurement team?"

Step 4: Enrich by Tier

Tier	Action	Tools	Time
🔴 Chain	Website deep + LinkedIn + news search + chain procurement mapping	agent-browser + agent-reach (Exa)	3-5min each
🟡 Mid	Website basics + FB	agent-browser	1-2min each
🟢 Small	Skip — Maps data sufficient	—	0

Chain enrichment with agent-browser:

1. agent-browser open "[website URL]"
2. agent-browser snapshot -i → extract Services, About, Staff, Contact
3. Check for Portfolio/Cases and News/Blog pages for expansion signals
4. For chains: Look for corporate/region procurement contacts, preferred vendor programs, and expansion news

Chain news search with agent-reach:

mcporter call 'exa.web_search_exa(query: "[company name] expansion OR new location OR equipment", numResults: 5)'

Chain procurement mapping (chains only) — See references/chain-strategy.md for full approach:

• Identify: local manager → regional operations manager → VP of operations / procurement director
• Sources: LinkedIn, corporate website "careers" or "partners" page, news about leadership changes
• Goal: find the person who can approve equipment purchases for multiple locations

Step 5: Score & Rank

Match each candidate against the profile card:

Factor	Rule	Points
Buy signal	Expansion / new service / new equipment	+5 (strong) / +3 (medium) / +1 (weak)
Industry match	Business type matches profile	+3
Scale match	Review count / bays similar to profile	+2
Service overlap	Same services as profile	+2
Geo similarity	Similar area type	+1
Business age	Similar years in operation	+1
Chain multiplier	Chain store (multiple locations = bulk potential)	+3
EV/high-end certification	EV Certified / LUXE / premium line	+4

Tie-breaking: buy signal strength → chain (bulk potential) → has phone → closer scale match

Total score	Priority	Action
10+	🔴 High	Call within 48h
6-9	🟡 Medium	Call this week
<5	🟢 Low	Call when available

Step 6: Generate Custom Sales Openers

Not templates — custom for each prospect based on their data.

Opener must accomplish 3 things: (1) prove you know them, (2) state your purpose, (3) invite dialogue.

Data source	How to use in opener
Buy signal	"Saw you just added [service related to your product]"
Similar customer	"We supplied [product] to [similar customer] in your area"
Business type	"Since you do [their business type]..."
Key clues	"As an [industry certification] shop..." / "Working with [their key client]..."
Tier	High→emphasize quality & custom, Mid→value, Low→entry-level
Chain store	Key opener question: "Is equipment purchasing handled locally, or should I speak with your regional/corporate procurement team?"
Premium/certified line	Reference their specialization: "As an EV-certified shop, you need [specific configuration] — we've done those."

Step 7: Output (3-layer structure)

Save to prospect-data/{batch}/:

prospect-data/{area}-{date}/
├── index.json          ← Lightweight index, instant search
├── P001.json           ← Full detail for first prospect
├── P002.json           ← Full detail for next prospect
└── call-list.csv       ← 11-column CSV for calling

See examples/ for sample output files.

Then export CSV from index + P###.json files for calling.

index.json — Search/filter only (few KB):

{
  "batch_id": "orange-ca-2026-05-19",
  "source_customer": "ABC Auto Body",
  "generated": "2026-05-19",
  "search_areas": ["Orange CA"],
  "product": "Customizable per industry",
  "chain_strategy": "Chain stores included — call local first to identify procurement decision chain, then escalate to regional/corporate",
  "prospects": {
    "P001": {
      "name": "Bob's Auto Body",
      "city": "Orange CA",
      "priority": "高",
      "tier": "中高端-独立",
      "status": "待联系",
      "tags": ["[industry]", "[business type]"],
      "file": "P001.json"
    },
    "P013": {
      "name": "Crash Champions Orange",
      "city": "Orange CA",
      "priority": "高",
      "tier": "连锁-中高端",
      "status": "待联系",
      "tags": ["collision", "chain", "Crash Champions"],
      "file": "P013.json"
    }
  }
}

P001.json — Full detail (all collected data + contact log):

{
  "id": "P001",
  "name": "Bob's Auto Body",
  "phone": "[EXTRACTED_FROM_MAPS]",
  "city": "Orange CA",
  "tier": "Mid-high-Independent",
  "priority": "High",
  "buy_signal": "Added new [service]",
  "similar_customer": "Customer A",
  "business_type": "[industry service type]",
  "key_clues": "[specific observations from data]",
  "email": "bob@bobscorp.com",
  "chain_brand": null,
  "opener": "We supplied [product] to [similar customer] in your area — saw you recently added [service]. What [product type] are you currently using?",
  "status": "Pending",
  "contact_log": [],
  "tags": ["[industry]", "[business type]", "[certification]"],
  "maps_url": "https://maps.google.com/...",
  "rating": 4.5,
  "reviews_count": 87,
  "has_website": true,
  "website_url": "https://bobscorp.com",
  "raw_notes": "Reviews mention...",
  "source_customer": "Customer A"
}

P013.json — Chain store example:

{
  "id": "P013",
  "name": "[Chain Brand] [City]",
  "phone": "[EXTRACTED_FROM_MAPS]",
  "city": "Orange CA",
  "tier": "Chain-Mid-high",
  "priority": "High",
  "buy_signal": "National chain with stable equipment needs across locations",
  "similar_customer": "Customer A",
  "business_type": "[Industry] Chain",
  "key_clues": "[Chain brand] national chain + [city] location + online booking",
  "email": "",
  "chain_brand": "[Chain Brand]",
  "opener": "Hi, I'm with [company] — we manufacture [product]. [Chain brand] has a location here, and I'd like to learn about your equipment purchasing process. Is that handled locally, or should I speak with your regional/corporate procurement team?",
  "status": "Pending",
  "contact_log": [],
  "tags": ["[industry]", "chain", "[chain brand]", "online booking"],
  "maps_url": "https://maps.google.com/...",
  "rating": 4.6,
  "reviews_count": 120,
  "has_website": true,
  "website_url": "https://www.chainbrand.com",
  "raw_notes": "National chain. Key question: local manager vs regional purchasing.",
  "source_customer": "Customer A"
}

CSV export — 11 columns, ready to call:

优先级,店名,电话,城市,档位,购买信号,相似客户,业务类型,关键线索,邮箱,开场白

CSV columns map 1:1 to P###.json fields (priority→tier, etc.). CSV is a projection of the JSON, not a separate data source.

Status tracking (in P###.json, not CSV):

待联系 → 已联系 → 意向 / 无意向 / 回访中
                 ↘ 无人接听 → 再试

Step 8: Update contact status

When user reports call results, update P###.json:

"contact_log": [
  {"date": "2026-05-20", "action": "电话", "result": "无人接听", "next": "明后天再试"}
]

And update index.json status field accordingly.

Re-export CSV filtered by status when user needs a new call list.

Critical Rules

1. Every step must execute — skip only if data source has nothing (no website = skip website enrichment)
2. Review sampling, not all — use tiered sampling + keyword filtering per profiling reference
3. Social media by tier only — 🔴 chain gets full search, 🟢 small gets nothing
4. Opener is custom — never use generic templates, always tailor to prospect's specific data
5. Output is 3-layer — index.json for search, P###.json for detail, CSV for calling
6. CSV is a projection — all data lives in JSON; CSV is just 11 columns exported on demand
7. Chain stores ARE valid prospects — do NOT exclude them. Include with a different strategy: local call first → identify procurement decision chain → escalate to regional/corporate buyer. One chain deal can equal many independent deals.
8. Tier labels include chain distinction — use "独立" (independent) or "连锁" (chain) suffix in tier: e.g., "中高端-独立", "连锁-中高端"
9. Chain opener must ask about procurement — "Is equipment purchasing handled locally, or should I speak with your regional/corporate procurement team?"
10. Specialized/certified prospects are high priority — certifications (EV, ISO, specific industry standards) indicate higher equipment requirements and justify premium positioning
11. DATA INTEGRITY — NO FABRICATION — All data in outputs MUST come from actual agent-browser searches, web_fetch calls, or other real data sources. NEVER invent, infer, or hallucinate business details. If a field cannot be verified from real data, mark it as "unknown", "not found", or "pending verification". If a search returns no results or fails due to network issues, report this honestly to the user instead of generating placeholder data.
12. TRANSPARENCY ON DATA GAPS — If Google Maps returns restricted view (limited details), if agent-browser fails to load, or if a business has no visible phone/address/rating, document this in raw_notes and adjust the priority accordingly. Do not fill gaps with assumptions.
13. VERIFICATION REQUIRED — Before marking any prospect as "ready to call", confirm that the phone number was actually extracted from a live page (not a template). If the number is a placeholder or unverified, flag it explicitly: "phone_status": "unverified_placeholder".
14. PLACEHOLDER PHONE POLICY — Output phone numbers must come from real page extraction. Any 555, 0000, 1234, or similar placeholder pattern must be marked "phone_status": "unverified_placeholder" and must not be written to the call-list.csv dial column.
15. OUTREACH BOUNDARY — The generated call list is for business-to-business contact only. Honor "do not call" requests. Stop calling a prospect upon request and mark status as "do_not_contact". Identify caller, company, and purpose on every call.
16. DATA RETENTION — Delete or archive prospect files when the campaign ends. Do not share raw JSON/CSV with third parties without consent.
17. SELF-OPTIMIZATION IS MANDATORY — The skill MUST automatically detect coverage gaps and adjust keywords/centers between search passes. Do NOT require user feedback to fix zero-result keywords or missing chain brands. Document all auto-adjustments in candidates-raw.txt with [PASS-N-ANALYSIS] markers.
18. COVERAGE GAP REPORTING — After all passes complete, generate a coverage-report.json documenting: (a) which center+keyword combinations returned zero results, (b) which auto-adjustments were applied, (c) estimated coverage percentage of target businesses in the area. This helps users assess completeness without manual review.
19. SUBURBAN EXPANSION IS AUTOMATIC — For cities >2M population, if initial 4-6 centers yield <50 unique prospects, auto-expand to 8-10 centers covering satellite cities. For Houston-sized markets, target 80-100+ prospects before stopping. Do NOT require user to ask for "more coverage" — the skill should self-assess and expand.
20. FILTER CALIBRATION IS REQUIRED — After first dedup, if >20% of valid prospects were filtered out (e.g., chain brands, legitimate shops with unconventional names), auto-relax filter rules and re-run. Log filter false-positives in candidates-raw.txt.
21. CHAIN BRAND LEARNING — Maintain a dynamic chain-brands-detected.json per search session. If a brand appears in >2 locations, add it to the known chain list for future searches. Share learnings across sessions via MEMORY.md updates.

Version History

• v2.0.0 (2026-05-23): Added self-optimization protocol, coverage gap reporting, chain brand learning, suburban expansion rules, and filter calibration. Houston field test validated 90 prospects from 11 centers with 10 auto-adjustments.
• v1.0.0 (2026-05-19): Initial release with multi-center search, chain strategy, 3-layer output, and custom sales openers.

---

End of SKILL.md