Resources

Implementation Guide / Customs & logistics

From incoming documents to a checked system record

Plan document extraction, field checks, duplicate handling and system updates before connecting an inbox to live operations.

Concept photograph of a technical professional connecting software and checking system information

The work continues after the document is read

Customs and logistics teams receive invoices, packing lists and other business documents, then organise information into the relevant records and check for gaps. A field-extraction demonstration is only one part of that work. In daily use, staff also need a way to handle revisions, incorrect associations and failed system updates.

Limit a first release to one document source, one document type and one business system. Describe the complete path from receipt through extraction, review and entry before expanding formats. In this scope, document processing means organising and checking information; staff with the relevant responsibilities confirm the final declaration content.

Define the fields before extracting them

Ask staff to demonstrate real documents. Which fields must be entered? Which come from another system? Which require calculation or judgement? Similar labels can mean different things across supplier documents.

Use a field map to agree the working rules:

Field Check If unclear
Business reference Correct job or transaction Ask the owner to confirm the link
Date Date type and format Keep the source text for review
Amount and currency Value, currency and totals Hold the update and show differences
Quantity and unit Cartons, items, weight or other unit Review the document and related record
Customer or supplier Existing party matched correctly Avoid creating an uncertain duplicate

Keep the source document and relevant location alongside extracted fields. A reviewer should not have to search the whole document pack for one number. An uncertain field should remain visible rather than being filled with a plausible-looking value.

Separate extraction, checks and business decisions

Extraction finds information in a document. Checks examine required fields, formats and associations. Business decisions may concern interpretation, declaration handling or special circumstances. Separating these layers makes it easier to assign rules and staff responsibility.

A clearly read document can still be attached to the wrong job. A model’s confidence can inform review, but does not replace the business check. The business owner should decide which fields and actions require confirmation.

Begin with a reviewable draft and its source shown together. Record why staff make corrections. This reveals whether a problem comes from document quality, field definitions or matching rules, rather than reducing every issue to one overall accuracy figure.

Plan for duplicates and revisions

The same email may be forwarded. A document may return under another name or as a revised version. Keep version, source and business associations so staff can distinguish a duplicate from an addition or an update.

Show changes before overwriting reviewed information. When documents conflict, collect the differences in a review item and assign someone to resolve them. A retried system update should also check whether it already created the target record, so that retrying does not create a second transaction.

Make connection failures visible

Before quoting integration, confirm how the existing software can be accessed, which actions it supports and who can authorise access. If an appropriate interface is unavailable, file import or a review queue may be a useful first step. Fully automatic entry should not be promised before that check.

A connected workflow should expose the source document, processing state, target record and failure reason. Staff need to distinguish information written to the system from information only prepared for review or waiting for retry. Put a defined limit around retrying and assign unresolved items to a person.

Test a batch where some records succeed and others fail. Running the entire batch again may duplicate successful entries. The team also needs a way to keep working while a connection is paused.

Accept the workflow against representative samples

Include clear documents, poor scans, missing pages, absent fields, different units, revisions, duplicates and a document with no matching business record. Define expected fields, issues to flag and actions requiring staff confirmation for each sample.

Measure field correctness, exception detection, review time and the correctness of the final business record separately. Include corrections and failed-update handling in any time-saving calculation. That is the basis for judging the complete workflow, rather than the extraction step alone.

The logistics project case describes related work. Use the testing and handover guide to plan acceptance testing for your own workflow.