Why Unlabelled SKUs Kill Warehouse Throughput

Published 2026-09-05 by Faraz Rahimi

Missing barcodes do not just slow a scan. They turn a 4-second pick into a 4-minute scavenger hunt. Here is the operational cost and the lookup pattern I use instead.

Tags: Computer Vision, Warehouse, Operations, Python, SKU, unlisted

A barcode that is missing, crushed, or wrapped in stretch film is not a data-quality issue on a dashboard. It is a person standing in an aisle with a cart behind them, guessing.

I have watched that guess turn into a radio call, then a walk to a computer, then a Ctrl-F through a spreadsheet that is already two days stale. By the time the SKU is found, the original pick wave is cold.

The expensive part is not the lookup. It is the context switch. Operators drop the physical problem, solve an information problem, then try to rebuild the physical problem in their head.

What I measure instead of scans per hour

Throughput metrics hide unlabelled inventory because the scan never happens. I track time-to-identity: camera up to SKU on screen. If that number is over a couple of seconds, people stop trusting the tool and go back to asking a lead.

I also track how often identity is confirmed by dwell — someone opening the record and staying — versus rejected with "not this item." Rejections are gold. They are labelled hard negatives you did not have to sit down and annotate.

A lookup path that does not require a perfect barcode

Photo in, candidates out, location attached. The model only has to be good enough to put the right item in the top three. A human can tap the right card faster than they can spell a SKU into a keyboard with gloves on.

Then a SQL query on last-scan location beats tribal knowledge. "It is usually on the left" is not a bin.

Do not replace barcodes

When a barcode exists and scans, use it. Computer vision is the fallback for the ugly 8% that creates 80% of the yelling. If you try to replace a working scan process, you will lose the floor on day one.

Original post: https://farazrahimi.com/posts/why-unlabelled-skus-kill-warehouse-throughput