How Portaxo learns the customs name for your product

Customs entries are written by brokers, not by marketers. Search the phrase a shopper would type and you often get the wrong companies — not fewer companies, different ones.

That gap is the reason Portaxo (portaxo.com) treats wording as a first-class step, not a search tip. You describe a product in plain language; the system works out how US import filings actually name it, then ranks importers against that learned phrasing. This post is the technical version of that step.

For the manual method — reading competitor filings and collecting variants by hand — see what customs calls your product.

Broker language vs marketing language

A seller says “tape measure.” A broker filling a bill of lading at speed may file MEASURING TAPE, TAPE MEASURE, HAND TOOL, or a longer SKU string that happens to include one of those phrases. The English is close enough that a human shrugs. The database does not shrug: it matches tokens. Near-synonyms can point at different consignees, different volume leaders, and different noise (freight forwarders vs manufacturers).

Portaxo’s job is not to guess the “correct English.” It is to find the wording the shipment corpus answers to for that product intent.

Propose, then let the data vote

The learning loop is deliberately cheap and measurable:

1. A model proposes candidate phrasings from the plain-language brief — short, blunt variants a broker might actually type, not catalog copy.
2. Each candidate is tried against US import records. Hits, top consignees, and shipment volume are compared — not vibes, counts.
3. The wording the data answers to wins. That winner becomes the category’s working query for ranking buyers (and, when you switch direction, suppliers).
4. Specific variants stay separate when they mean a different product. A phrase that pulls a different market is not collapsed into a synonym just because the words look related.

The model suggests. The customs data picks. That order matters: if the model picked alone, you would get fluent wrong answers.

One measured example: tape measure → measuring tape

Same product. Two wordings. Different top of list.

The second phrasing is not “smarter English.” It is the string the filings in this category respond to with real importers and usable volume. Searching the shopper phrase did not return an empty set — it returned the wrong companies with enough confidence to look convincing. That is worse than empty.

When Portaxo learns a category, it keeps the winner that behaves like the right-hand row: manufacturers and active importers, not the logistics name that happens to touch the category.

Caching: the second search is free

Learning a category costs work (and credits) once. The resolved wording and the discovered company set are cached for that product category so the next search — yours or another teammate’s — can reuse the result instead of re-probing every synonym.

Practical consequences:

• First search pays for discovery and wording resolution.
• Repeat search on the same learned category is instant and does not re-bill empty synonym probes.
• Follow-ups in the AI Agent (size filters, “only the small ones”) reuse the paid result set rather than starting a new category from scratch.

Empty searches stay free: if a candidate wording matches nothing, you are not charged for the miss.

What this is not

• Not HS-code magic. Codes narrow a neighbourhood; wording still finds the street. See the HS discussion in what customs calls your product.
• Not a thesaurus. Synonyms that pull different markets stay split.
• Not a guarantee every broker uses one phrase. The cache stores what worked for this category in this corpus; new products and rare SKUs may need another learning pass.

Where to try it

Describe a product in Find buyers or ask the AI Agent in plain language. Portaxo resolves the customs wording, ranks US importers by fit and activity, and — when you approve — unlocks contacts in the same workspace.

Try Portaxo free — 30 credits a month, no card.