The challenge
Operations staff received commercial invoices, packing lists, bills of lading, and air waybills as PDF attachments through a shared email inbox. Each document was opened manually, reviewed, and re-keyed into the freight management system. Scanned documents required additional checking before data could be entered.
Supplier document layouts varied across the client base, which meant staff could not apply a consistent reading pattern. Shipment references were sometimes missing from individual documents, and verifying values across the multiple documents that belong to a single shipment added further time to each transaction.
The combination of repetitive data entry, layout inconsistency, missing references, and cross-document checking created a processing bottleneck. Baseline figures for document volume, handling time per document, and staff hours spent on this work have not yet been confirmed.
The approach
Netofficials began by examining the existing workflow and collecting sample documents representing the range of supplier layouts and document types the client receives. This sample set was used to understand variation in field placement, document quality, and the relationships between documents belonging to the same shipment.
The custom software development and AI work were designed together so that the classification model, the field extraction logic, and the validation rules all fed into a single workflow rather than operating as separate tools. The fields targeted for extraction were shipment reference, document type, invoice number and date, shipper, consignee, container number, package count, gross weight, currency, and invoice total, with the applicable fields varying by document type.
A human review queue was built into the design from the start. Records that had missing mandatory fields, low-confidence extraction results, duplicate document signals, or conflicting values across related documents were held for staff review rather than passed forward automatically. Staff could view the source document alongside the extracted values, make corrections, and approve the record.
The solution
The delivered system handles the full document intake process from the shared email inbox through to records ready for the freight management system. Incoming PDF attachments are ingested automatically, classified by document type, and processed for field extraction.
Extracted records that pass validation are prepared for transfer into the freight management system. Records that fail validation thresholds enter the review queue, where staff check the source document, correct any errors, and approve the record before it moves forward.
The custom software development work produced the intake pipeline, the review interface, and the integration layer connecting the system to the freight management platform. The AI and machine learning work produced the document classification model and the field extraction capability that handles varied supplier layouts. The specific OCR provider, AI model, backend framework, database, hosting environment, and the mechanism used to transfer approved records into the freight management system have not yet been confirmed.
The results
The system removed the step of manually reading and re-keying each document from the shared inbox. Staff time on freight document processing shifted from full data entry on every document to reviewing and correcting only those records the system could not complete with sufficient confidence.
Validated metrics covering handling time per document, the proportion of records requiring correction, the review rate, and total staff hours over a defined pilot period have not yet been supplied. Those figures will be added once the pilot measurement period is complete.
