When a Kuwait contract needs a clause extract, not a chat

The job is a signed file they already keep. A Kuwait ops desk has the vendor PDF. Someone still highlights the end date, the notice window, and the cap, then types those three lines into the sheet they already open on Monday.

Teams ask for “document AI.” That is a shelf. The job is named fields from a contract they already signed, landing as a row they already trust — not a chat that rereads the whole pack.

What the extract actually does

The fields are already named. Counterparty, start, end, notice days, a currency cap if the page prints one. Text generation here means a structured read of that file, written onto the sheet or folder they already use, then a stop. A miss stays blank. A person can open last week’s row and see the file, the extract, and what was kept.

Policy PDF annotation is the same shape on a denser form: name each input, keep the geometry, refuse a field the page does not show. This job is thinner. It does not map every checkbox. It does not fill a blank.

This is not writing a known record into a mapped form. Fill spends a row you already have. Extract creates the row. It is also not posting a supplier bill in the ERP. A bill is one total onto a ledger. A contract is a handful of clauses the desk already tracks by hand.

When we leave the contract with a person

If two revisions sit in the same folder, software will quote the wrong end date. If a lawyer still has to interpret a sentence before anyone may rely on it, keep the person on the row. If the scan has no text layer, the first slice is a readable file, not a chat box.

Desks often keep Arabic on the paper and English in the sheet. That is a copy rule on named fields. A KWD figure the page does not print must not appear on the draft.

Bring the last contract a person typed into the sheet and the three fields they actually copy. Contact us, write hello@jamilglobal.com, or book thirty minutes.