Case Study - Order entry reduced from 30+ minutes to seconds, fully automated.

Overview

XiuXiu Paper Art works with customers who send orders in a variety of formats — PDFs, Excel spreadsheets, CSV files, and Word documents. Each order contains product reference codes and quantities, but the format and column naming varies between customers. Staff had to manually open each file, read through every line, look up the matching product SKU in their catalog, and type each item into Holded (their ERP system) one by one. A single order could take 30 minutes or more, and mistakes in reference matching or quantity entry were common.

The automation starts with a simple Tally form. The team selects the customer from a dropdown, uploads the order file, and hits submit. Behind the scenes, OpenAI reads the uploaded file — regardless of format — and extracts every product reference code and quantity. Each extracted product is matched against a Google Sheet product catalog using both product name and barcode to find the correct Holded SKU.

If every product matches, the sales order is created in Holded within seconds with the correct contact and all line items populated. If any products can't be matched, the team receives an email listing exactly which items are missing from the catalog so they can update the product list and resubmit. The uploaded file is automatically deleted from OpenAI after processing for data privacy.

The system handles minor formatting variations in product codes — differences in casing, spacing, or column naming — because the AI reads the content semantically rather than relying on rigid parsing rules.

What we did