CASE SUMMARY
The valuable automation was not OCR alone, but the complete path from extraction and price matching to exception review and ERP output.
The client received purchase orders in multiple formats. Staff copied product, quantity, price and date fields, checked several quarterly price lists and rebuilt the data in an ERP Excel template.
The new system combines document recognition, price search, risk flags, human correction and formal export in one screen. Normal lines process automatically, while similar product codes, blank prices and ambiguous numbers go to staff for a decision.
Each order required repeated switching between a PDF, several spreadsheets and an ERP template.
Documents could be searchable PDFs, scanned PDFs or images. Product codes could contain leading zeros, punctuation or version suffixes. After transcription, staff still searched price lists for approved prices, units, minimum quantities and lead times.
European number formats were a particular risk. 1.000 may mean one thousand units, while 2.172,90 means 2172.90. Text extraction without field and total context can create quantity or price errors.
The system selects text or visual recognition based on the file, then validates results with transaction logic.
After upload, the system extracts the PO number, date, line number, product code, description, quantity, unit, price, total and requested delivery date. It uses text when reliable and visual recognition for scanned files.
Quantity, unit price and line total are cross-checked. If they do not reconcile, the interface asks the user to verify the source instead of silently accepting a plausible-looking result.
All active price lists are searched together, with limited candidates shown only when an exact match is absent.
Three price-data families contain nearly three thousand records. Different workbooks and sheets are normalised into one structure, so users do not need to choose a spreadsheet first. Every match retains its workbook, sheet and row source.
When an exact code is absent, the system shows only a small number of similar candidates and preserves the original code. A candidate never replaces the PO automatically, reducing the risk that fuzzy matching selects another valid product.
Blank price
Flagged centrally for supplier or internal confirmation.
Duplicate code
All relevant sources are shown instead of selecting one silently.
Similar product
Candidates require a human selection and the decision remains traceable.
After review, the system creates an ERP-ready Excel file from the formal template.
The output preserves required sheets, column order and date formats. Approved quarterly price, unit and lead time come from active data, while expected delivery is calculated from order date and matched lead time.
Administrators can upload a new quarterly price list, review readable records, blanks and duplicates, then activate it to replace only that source. A quarterly update becomes an auditable data task rather than a code change.
Based on current order volume, the client estimated an 80% improvement and about 60 hours saved monthly.
Staff work shifted from cell-by-cell entry to exception confirmation. Source locations, original codes and human selections remain in the result for later review.
The figures are client estimates based on current workload. The next step is to record monthly order volume and old versus new handling and review time to create a durable baseline.
Frequently asked questions
Can the system process scanned PDFs?
Yes. It uses visual recognition when direct text is insufficient, while all extracted results enter the same review interface.
Are similar product codes matched automatically?
No. Only exact codes can match directly. Similar results are suggestions that require user selection.
Does the system replace final review?
No. Missing products, blank prices, duplicates and numeric inconsistencies are flagged, and staff confirm the result before formal export.
EXPERT AUTOMATION CONSULTATION
Does your team have a repetitive, error-prone workflow?
Tell us how the process works today, its monthly volume and the common exceptions. We will identify the part worth automating and define a testable first phase.