I have watched receiving teams re-type tracking numbers that were already printed two inches from their keyboard. That is not a staffing problem. It is a pipeline with a hole.
A small Python script that reads a scan or an OCR crop, validates the pattern, and writes the row beats a "digital transformation" slide.
Make failure loud
If the parse fails, the script should refuse to write a partial row. Silent NaNs become inventory ghosts. I pop a window with the image crop and one correction field.
Idempotency matters. Scanning the same label twice should update a timestamp, not duplicate a receipt.
Sit next to the printer, not in a repo nobody runs
The script has to live where receiving lives: a shortcut, a hotkey, a scanner wedge. A GitHub README is not an interface. I log every run so when it breaks at 6am I can see the last good label.
