One shipment, six documents, one set of numbers

Ask someone outside the trade what a customs broker does and they will describe filing. Ask a broker and they will describe checking.

One shipment arrives as a small pile of paper written by different people in different countries for different purposes. A commercial invoice from the seller. A packing list, often from the same seller but produced by a different department on a different day. A bill of lading or air waybill from the carrier. A certificate of origin. Possibly a licence, a fumigation certificate, an insurance document, a proforma that has since been superseded.

Each is internally consistent. The job is that they must be consistent with each other before anything is declared, and they frequently are not.

The disagreements that cost you

They are boringly consistent across trades:

• Value. The invoice total, the sum of the line items, and the value on the certificate of origin. Discount lines and freight terms are where these part company.
• Weight. Gross versus net, on the packing list versus the bill of lading. One is in kilos and one is in pounds more often than anyone would like.
• Quantity. Cartons versus pieces versus pallets, counted differently on two documents that both say "quantity".
• Party names. The seller on the invoice, the shipper on the bill of lading and the exporter on the certificate of origin are frequently three renderings of one company, and occasionally three different companies for reasons that matter.
• Description of goods. The words that drive classification, and the point where a vague invoice line becomes your problem rather than the seller's.

Catch these before filing and it is a phone call. Catch them after and it is an amendment, a delay, a query, or a penalty — and the cost is never only the fee.

What a table does here

The reason this work resists automation is that it is not extraction from a document. It is comparison across documents that were never designed to be compared. There is no shared key, no schema, and no guarantee the same fact is even called the same thing twice.

That is the shape a document grid fits. One row per document in the shipment file, one column per fact that has to agree, and the disagreements become visually obvious because they sit in the same column with different values in them. You are no longer holding six documents in your head — you are reading one table and looking down columns.

Ragextract does not ship a customs table, so the columns are ones you write yourself. In practice that is a short list of questions in plain language — "What is the total invoice value, including currency?", "What is the gross weight, and in what unit?", "Who is named as the exporter or shipper?" — asked of every document in the set. Columns are prompts, so the phrasing is yours and you can tighten it as you learn what your suppliers' paperwork does. Once it works, save it as a table you reuse per shipment.

The second thing that matters here is provenance. Every answer carries the quote and page it came from, so when the invoice says one weight and the bill of lading says another, you are not adjudicating between two numbers a system produced — you are looking at two lines of two documents, and deciding which one is right is a job you already know how to do.

Getting the file in

Shipment documents rarely arrive as a tidy folder. They arrive by email, in a chain, as attachments from three parties over two days.

Two ways in are worth knowing. You can forward the email — attachments become documents in a table without anyone opening a browser, which suits a workflow where the paperwork is already landing in an inbox. Or you can connect the folder where your files end up, if that is a shared drive or SharePoint site, and let new documents arrive on their own. Both are tier-gated features rather than part of the free grant, so check what your plan carries before building a process on them.

What it will not do

Worth being blunt, because the over-claim in this space is obvious and common.

It does not classify. No HS code, no duty rate, no tariff engine. It will read a description of goods and put it in a cell; deciding what that description is remains the part of the job you are licensed for.

It does not file anything. There is no connection to a customs system, no entry submission, and no status to track.

It does not know the rules. It has no view on whether a certificate is required for this origin, whether a licence has expired, or whether a preference claim is valid. It reads documents; it does not consult regulations, and it will not go and look one up.

It will not fetch anything it wasn't given. The engine reads the files in front of it and nothing else — no registry lookups, no web search. That is what keeps every cell traceable to a page, and it means a fact absent from the paperwork is absent from the table rather than guessed.

Where to start

Take one shipment file you have already cleared — ideally one that had a discrepancy you caught late. Put every document from it in one table, write six columns for the facts that had to agree, and see whether the discrepancy shows up as two different values in one column.

If it does, you have a check that runs before filing instead of after.

• Writing columns as questions
• Bundling several documents into one row
• Citations, and what they let you verify