Customs Lead SOP（海关数据开发客户）

Portable Agent Skill · {baseDir} = this skill folder
Companion QA skill: sibling customs-pitfalls (same skills/ parent)

Goal

Answer three questions with public trade data:

1. Which markets are buying? (Type B — national / HS stats)
2. Which companies have actually imported? (Type A — firm-level BOL, or secondary leads where A is closed)
3. Is there a supply-chain entry window now?

Customs data is a lead signal, not a closed deal. Conversion still depends on research, contacts, value prop, and follow-up.

Before every delivery, run the checklist in sibling skill customs-pitfalls.

Runtime notes

Tools (any available): web search, web fetch, browser, read local CSV/Excel. No paid API required for the methodology.

Data types (pick before querying)

Hard split: Step 2 uses B; Step 4 needs A (or paid A export / secondary leads). Never invent firm names from B.

Tool map: references/tools.md

Workflow (execute in order)

Task Progress:
- [ ] 0. Collect inputs (+ optional CSV/Excel exports)
- [ ] 1. Confirm HS / HTS / CN codes
- [ ] 2. Rank markets with Type B stats (Priority / Opportunity / Caution)
- [ ] 3. Check tariff barriers
- [ ] 4. Find real buyers (Type A where public; else secondary leads / user export)
- [ ] 5. Score & tier buyers
- [ ] 6. Find contacts + translate evidence into outreach drafts
- [ ] 7. Write report to disk
- [ ] 8. QA with customs-pitfalls

Step 0 — Inputs

Step 1 — HS codes

Do not treat AI-guessed codes from Chinese names as final.

Prefer (in order): historical export declarations → customer clearance docs → forwarder/broker records → comparable filed shipments → destination official tariff lookup.

Output: hs_codes[] + confidence (confirmed | candidate) + source.
If unconfirmed, tag later market/tariff analysis with HS risk.

Step 2 — Market priority (Type B only)

Use stats tools (China Customs, UN Comtrade, USITC DataWeb, Eurostat Comext, Canada CIMT).

Compare: import value, 3y growth, imports from China, China share, monthly seasonality, rough unit value.

Split Europe by country. NL/BE hub imports ≠ end consumption.
Comtrade output stops here — it never feeds Step 4 company rows.

Step 3 — Tariffs

WTO Tariff & Trade Data: MFN Applied / Bound, etc. Put size and duty in one table before recommending entry.

Order-level truth = latest national tariff + FTAs + AD/CVD + broker advice. This step is screening only.

Step 4 — Real buyers

4a — Where public firm-level ocean BOL is usable

US reverse lookup:

1. Competitor English name → importers with ≥2 shipments in last 12 months
2. Or diligence known accounts: continuity, origin countries, supplier concentration, product fit, arrivals in last 6 months
3. Cross-check: address, name variants, affiliates, product text
4. Drop / flag: To Order, Bank, masked consignee, forwarder, customs broker (see customs-pitfalls)

4b — UK / EU / most markets without public firm BOL

Due to privacy regimes (e.g. GDPR), most UK/EU official sources no longer publish US-style shipment-level consignee manifests. Do not treat UK Trade Info (or similar) as ImportYeti-for-Europe.

For these markets:

1. Finish Step 2–3 ranking with Type B stats
2. Build company pool via secondary leads: trade shows / exhibitor lists, industry directories, LinkedIn, association members, news of openings/expansions, optional sibling B2B skills
3. If the user has a paid customs platform covering that market, ask for export CSV/Excel and run 4c
4. Label evidence as secondary_lead or paid_export, never fake “BOL confirmed”

A few jurisdictions historically publish more open firm-level trade data (coverage changes — verify before claiming). Prefer stating limits over guessing.

4c — Manual export handoff (anti-bot / login / paid tools)

When browser/fetch hits Cloudflare, login walls, captchas, or free-tier caps:

1. Stop scraping; do not invent buyers
2. Ask the user to export CSV/Excel from ImportYeti / paid library / internal dump
3. Accept bol_export_files (or pasted tables) → parse → dedupe → filter pitfalls → score → draft
4. Note in the report: source=user_export + filename + row count + date

This path is the preferred production path for teams that already pay for customs software: Skill owns cleaning, scoring, and drafts; the platform owns raw data.

Step 5 — Score buyers

Full rubric: references/scorecard.md

Tiers: ≥80 1:1 focus · 60–79 normal follow-up · 40–59 nurture · <40 deprioritize

Secondary leads without shipment proof: cap confidence; do not invent frequency/recency scores from thin air — mark evidence_thin.

Step 6 — Contacts & outreach (translate, don’t expose)

Customs data answers “who buys”, not “who to email”. Roles: Purchasing / Buyer / Sourcing / Category / Supply Chain / Product Manager / Owner / MD.

Hard rules:

• Never write: “We found you from customs data / bill of lading / ImportYeti…”
• Never cite container counts, named Chinese shippers, or BOL IDs in customer copy
• Translate internal fields → product / replenishment / capability observations (see pitfalls Pitfall 12)
• Keep provenance in CRM / report only

Angle by buyer type: see examples/humanoid-harmonic-reducer.md (ToB) and examples/pet-products.md (consumer)

Step 7 — Report output

Write under output_dir (default ./hunt_result/).

Filename: customs-YY-MM-DD-{slug}.md (append -2 on same-day collision). Optional .jsonl / .csv of scored buyers.

# Customs Lead SOP — {product} / {markets}

## 1. HS codes
## 2. Market tiers + key metrics (Type B sources)
## 3. Tariff / access summary
## 4. Buyer list (source type A/export/secondary, evidence, score, tier, flags)
## 5. Outreach order + translated angles
## 6. Draft emails (optional, high tier only)
## 7. Risks & data limits (customs-pitfalls)
## 8. Next three actions

Each buyer needs: source_tools or source=user_export + time window + product/HS evidence + supplier-structure signal if any + flags (possible_agent_or_3pl, possible_customs_broker, consignee_masked_or_to_order, possible_reexport_hub, hs_risk, evidence_thin, …).

Optional handoffs

If these skills exist in the host workspace, you may chain them after scoring:

Compliance

• Public trade/BOL/stats + public web only (plus user-owned exports they already have rights to use)
• Drafts only — never auto-send email
• No fake “official customs buyer list” claims
• Free tiers are sample-limited; anti-bot is normal — hand off to CSV rather than bypass walls
• Do not instruct bypassing login walls or captchas

Minimum start pack (same day)

1. Confirm one accurate HS from filing docs
2. Compare three target countries on UN Comtrade (macro only)
3. US: score ~20 recent buyers from ImportYeti/ImportInfo or user CSV; other markets: secondary leads + clear “no public BOL” caveat