Automating a daily receiving check: 17 of 42 removals were wrong

Another piece of real work from Food Service Direct – Unilever USA. Every day a supplier delivery arrived, and someone had to decide, line by line, what to do with every item that hadn't been scanned into the warehouse: remove the purchase order, cancel it, or leave it for review.

The manual routine

It was a daily spreadsheet job: download the files, sort them, look each line up against the delivery file, check stock, and decide. It was slow, and the rules lived in one person's head.

What runs now

• A scheduled job downloads the supplier's confirmation and delivery files every day.
• It checks every unscanned line against the order database.
• Each line is sorted into the right action: remove, cancel, or cancel as a special order.
• The team gets ready-to-upload files by email, plus a tab of partial scans for a human to review.

The business rules were gathered from the operations owner over several months and kept in a decision log, so every rule has a reason behind it. I used AI tools (Claude Code and Cursor) to draft the code faster, and checked every change before it went live.

The 17-of-42 finding

Once the check ran on real data every day, a pattern showed up: the old routine was removing purchase orders too early. On one day, 17 of the 42 removals were for items that were scanned in later. Waiting one extra delivery before removing an order stopped those false removals.

Built to be trusted

• Special orders, which have long lead times, are handled separately; checked across 8,300+ order lines in 2026.
• If an input file is missing or wrong, the job stops and alerts instead of sending bad files.
• Every run is logged, with a success or failure email.

What this means for you

If your team checks deliveries, stock or orders by hand every day, that's what my Inventory & Order Operations service automates, with alerts when something needs a human.