FLIGHT MANUAL — DEEP DIVE

An automated data entry workflow was built to eliminate the manual re-keying of data between the client's procurement system and its ERP. Previously, purchase orders received via email were printed, manually entered into the ERP, and then filed — a process that was slow, error-prone, and consumed significant clerical time.

The workflow uses optical character recognition to extract structured data from the purchase orders, validates the extracted data against the vendor master and the open PO file, and posts the transaction directly to the ERP. Exceptions — unrecognised vendors, unmatched POs, data that fails validation — are routed to a human reviewer.

The accuracy of the OCR extraction was the critical success factor. The system was trained on a large corpus of historical purchase orders, with human-in-the-loop validation during the training phase to improve the model's accuracy on the client's specific document formats. The system now achieves over 98% straight-through processing.

The workflow has eliminated a category of error that was previously invisible: transcription errors that were introduced during manual entry and then propagated through the ERP. By removing the manual step, the data in the ERP is now as accurate as the source document — and the source document is retained as evidence.