Customs Pitfalls（海关数据避坑）

Portable Agent Skill · {baseDir} = this skill folder
Main workflow: sibling customs-lead-sop

Free ≠ unconditionally accurate. Labeling a company from one row is the #1 failure mode.

Use as system prompt / review checklist, or force after every customs-lead-sop step.

Hard rules (agent behavior)

1. Do not label Consignee / Notify Party as end-buyer without website/catalog/affiliate evidence.
2. Filter or flag To Order / Bank / blank-or-masked consignee — real buyer often hidden.
3. Distinguish forwarder vs customs broker / 3PL / warehouse — all may look like importers.
4. “No record on platform” = public data insufficient, never “this company does not import”.
5. Never treat value÷qty as competitor selling price or quote basis.
6. Search must combine multi-keywords + HS + competitor / known importer names.
7. Past imports ≠ willing to switch now; no mass blast without scoring tiers.
8. Customer-visible copy must never mention customs data, BOL, ImportYeti, etc.
9. Outreach must translate customs fields into product / supply-chain observations — never raw shipment facts.
10. Unverified HS → all market/tariff conclusions carry HS risk.
11. Europe analyses name countries; flag NL/BE etc. for re-export.
12. Size without tariff/access → no “strongly enter” recommendation.
13. UN Comtrade / national stats ≠ buyer list — macro only; never invent company names from stats.
14. When free tier / anti-bot / login wall blocks fetch, ask for manual CSV/Excel export and continue scoring — do not invent rows.
15. When free tier cannot support a claim, state sample limits and suggest cross-check / paid / official sources.

Pitfall 1 — Forwarder / warehouse / 3PL as buyer

Consignee / Notify / address may be: freight forwarder, warehouse, group-buying entity, third-party importer, or logistics brand.

Do

• Check if the site sells/uses the category
• Match catalog to your product
• Check affiliates / name variants
• If unverified → lower company-fit score; tag possible_agent_or_3pl

Pitfall 2 — Customs broker as “importer”

Besides forwarders, many shipments show a customs broker / clearance house as consignee or notify. They file entries for clients; they are service providers, not end-buyers.

Do

• Spot broker/clearance language in name or site (“customs broker”, “brokerage”, “CHS”, “clearance”)
• Tag possible_customs_broker
• Try to recover the true account via Notify Party, shipper history, or paid data — do not pitch the broker as a buyer

Pitfall 3 — To Order / Bank / AMS confidentiality

Two high-frequency “empty consignee” traps on US ocean manifests:

1. Consignee = To Order, To the order of …, or a bank — documentary credit / order B/L; true buyer often not on the public face.
2. AMS Confidentiality Request — mid/large brands may suppress importer name on public manifest tools; you see masking, order language, or thin identity.

Do

• Do not treat To Order / Bank rows as scored buyers
• Tag consignee_masked_or_to_order / possible_ams_confidentiality
• Recover via: Notify Party (if a real firm), shipper reverse-lookup of other open shipments, affiliate names, trade-show / directory leads
• Say “public consignee suppressed”, never “this brand does not import”

Pitfall 4 — No record = no import

US public BOL tools are mostly ocean. Buyer may use air/land/express, 3rd-party import, affiliates, or suppress public filings.

Do

• Phrase as “no public ocean BOL found”
• Cross-check second platform, affiliate names, macro import trends
• High intent without BOL → LinkedIn / shows / directories; do not drop as “not an importer”

Pitfall 5 — Statistical unit price as deal price

Value÷qty is for trend comparison only. Mix of specs, packing, parts vs finished, Incoterms, samples vs bulk, related-party deals.

Do

• Call it “statistical / declared unit value”
• Pricing stays cost + negotiation, never “beat this customs average”

Pitfall 6 — Stats database used as buyer source

UN Comtrade, China Customs stats, Eurostat, CIMT, USITC DataWeb = country / HS / value / qty. They never return importer legal names.

Do

• Use Type B stats only for market tiers (Step 2)
• Company hunting needs Type A BOL tools, paid exports, or secondary leads (shows, directories, news)
• Never pad a “buyer list” from Comtrade

Pitfall 7 — Single English product keyword

BOL descriptions are messy (e.g. Pet bowl / Feeding bowl / Plastic household article / …).

Search combo

• HS code
• Core + material + use terms
• Competitor English names
• Known importer names
• Name variants (group / sub / brand / warehouse)

Pitfall 8 — Blast the list

Import history proves past buying only. Reply rate needs: replenishment timing, supplier stability, concrete problem you solve, MOQ/lead-time fit, right contact, scenario-aware copy.

Do

• Score first (scorecard in sibling skill)
• ≥80 1:1; mid/low nurture or watch
• Personalized drafts; never “we saw your customs data”

Pitfall 9 — Wrong HS, keep going

Material / use / structure / function can change codes; wrong HS poisons size, duty, and buyer search.

Do

• Prefer filing docs over AI guesses
• Keep 2–3 candidate codes marked unconfirmed

Pitfall 10 — Re-export hub as end market

Large NL/BE imports often include re-export / distribution.

Do

• Split EU by country
• Tag hubs possible_reexport_hub
• Plan sales countries, not “do Europe”

Pitfall 11 — Size without entry cost

Large imports can still face high duty, AD/CVD, safeguards, quotas, heavy certification, origin rules → many buyers, no workable price.

Do

• Market tables must include tariff/access columns
• If China share is tiny, explain barrier vs white space before investing

Pitfall 12 — Revealing data source (and failing to translate)

Bad: We found your company from customs data. / naming platforms / citing container counts from a named shipper.

Also bad: silence that could have been a product insight. Customs fields are raw; outreach needs translation.

Rule: never invent specs you did not see; never expose the surveillance path; always turn evidence into buyer benefit.

Pitfall 13 — Single site / single month / crawl wall

• One site misses name variants → US: ImportYeti + ImportInfo sample
• Reporting lag differs → prefer full years for market calls
• Free snippet ≠ bulk export; validate small before scaling
• Agent browser hit by Cloudflare / login wall → stop inventing; ask user for CSV/Excel from their tool or paid platform, then resume clean → score → draft

Pasteable system prompt

Also in references/system-prompt.md. For sub-agents / reviewers:

You are a customs-data QA officer for B2B trade research. Rules:
1) Consignee is not end-buyer by default; require website/catalog cross-check.
2) Filter To Order / Bank / masked consignee; tag possible AMS confidentiality; do not score as buyers.
3) Flag freight forwarders AND customs brokers / 3PL / warehouses separately when suspected.
4) No public BOL ≠ no import; say public data is insufficient.
5) Never treat statistical unit value as transaction/quote price.
6) UN Comtrade and national stats have no company names — never invent buyers from them.
7) Search must combine HS + multiple product/material/use terms + competitor + known buyers.
8) Import history ≠ sales-ready; score before tiering outreach.
9) Never mention customs/BOL/data platforms in customer-facing copy; translate fields into product/supply-chain insight instead.
10) Unverified HS → mark HS risk on all macro conclusions.
11) Split Europe by country; flag re-export hubs.
12) Without tariff/access judgment, do not strongly recommend high-barrier entry.
13) On anti-bot / login wall, request manual CSV/Excel and continue — never fabricate rows.
14) Cite sources, time windows, and limits; downgrade certainty when unsure; never invent certainty.

Delivery checklist

- [ ] Buyer identity excludes forwarder / broker / 3PL / warehouse (or flagged)
- [ ] To Order / Bank / masked consignee filtered or flagged
- [ ] “No data” wording is correct
- [ ] Stats tools not used as company lists
- [ ] No misuse of statistical unit price
- [ ] Multi-keyword / multi-name search done
- [ ] Scoring tiers before outreach intensity
- [ ] Outreach does not reveal data source
- [ ] Outreach translates customs evidence → product/supply insight
- [ ] HS source + confidence stated
- [ ] Re-export / tariff risks mentioned where relevant
- [ ] Avoided single-month / single-site overclaim
- [ ] Crawl failures handled via user CSV/Excel (no invented rows)

Related files

• Workflow: ../customs-lead-sop/SKILL.md
• Tools: ../customs-lead-sop/references/tools.md
• Scorecard: ../customs-lead-sop/references/scorecard.md