A receiving clerk once cost his company three days over one mistyped lot number. Every carton on the pallet wore a clean Code 128 label with the right number sitting there in the bars. The delivery note came in as a scanned PDF, though, so he read the digits off the printout and keyed them by hand, and two of them swapped places on the way in. Purchasing then spent the rest of the week chasing a discrepancy that didn’t exist, across two separate warehouses, before somebody finally pulled the original label and spotted the typo.
Nothing about that is unusual. The data got captured the instant the label rolled off the printer. What nobody had was a quick way to get it back out of the paperwork.
The Number’s On The Page. Good Luck Reaching It.
A barcode is a machine-readable way of writing down an identifier, nothing fancier than that. GS1, the non-profit that runs the standards, keeps a whole menagerie of them: UPC and EAN on retail packaging, ITF-14 on shipping cartons, GS1-128 when logistics people need a batch or expiry baked in, DataMatrix for parts too small to hold anything bigger. Aim a scanner at one and it gives up a product ID, a serial, a date.
The problem starts when the code lands as a picture instead of a live scan. A supplier emails an invoice. A carrier hands over a bill of lading. Someone photographs a label on their phone under flickering warehouse light. Every bar is intact, and every bar is stuck inside a file that has no intention of handing your inventory system a thing.
That part is about to get heavier. GS1 has told retailers and hospitals to be ready by 2027 to read 2D codes, the QR and DataMatrix squares that pack a lot number, an expiry and a serial into one mark instead of the plain stripes we all grew up scanning at the till. More data crammed onto every label. More of it buried in documents nobody has the hours to retype.
The Re-keying Tax
Spend an afternoon beside an operations lead and watch the time drain away. Open a scanned goods-received note. Find the digits under the barcode. Type them into a field. Next document. A few hundred rounds of that a week is a salaried person doing nothing but ferrying numbers out of a picture and into a database.
People are unreliable at it, too, and predictably so. Hand-keyed data tends to carry roughly one error for every few hundred characters typed; a scanner misfires nowhere near that often. One transposed digit in a serial sends a warranty return to Memphis when it was meant for Reno, or throws a counterfeit flag on stock that’s perfectly legitimate. A barcode exists to make the chain of custody tighter in the first place, and inventory security comes apart the second the number in your log and the number on the box stop agreeing.
Read The Code Out Of The File Itself
So don’t print it. The barcode already lives inside the PDF or the image, fully formed and perfectly legible to a machine. There’s no reason to route it back to a loading dock and scan it off a carton a second time. Decode it where it sits.
That’s the job barcode text extraction does: it finds the symbol inside a document, reads it, and hands back the encoded value as text your systems can store, search and reconcile.
After that, most of the work stops being work. The receiving log populates itself. A serial gets checked against its purchase order the moment the file arrives, not three days later when someone opens a dispute. That clerk from the opening never has to squint at a printout again.
A Read You Can Trust, Or None Of It Holds
Decode accuracy is the hinge the whole thing swings on. The bars stand in for a number, the number points at a database row, and if it comes back even slightly off, the record it drags up is worse than no record at all. NetSuite makes the case that barcodes earned their place in supply chains precisely by stamping out the hand-entry mistakes we’ve been circling this whole time. Pull the data automatically and you only keep that win if the read is as clean as a scan gun’s.
Real documents don’t cooperate, though. Labels turn up rotated a quarter turn, faxed into grey mush, shrunk to a thumbnail, or crammed three to a page next to a coffee ring. Good tooling barely notices skew or a rough print run. Plenty of tooling folds at the first wrinkle. If I hand you one rule, make it this: benchmark on your ugliest inputs, the crumpled ones from the supplier you’re always complaining about, because those are what decide whether any of this survives your actual mailroom.
It Bolts Onto What You Already Run
None of this means tearing out your stack. Extraction is one step inside a flow you already have. Document arrives, value drops out, value flows into whatever you’re running today. A lot of smaller shops lean on logistics management software to track stock and push orders out the door, and extraction sits one notch upstream of it, so paperwork shows up already digitised rather than as a stack for somebody to key in.
If you keep one idea from all of this, keep the reframe. An inbound document is a data source, and your software should be the first thing that opens it. That sounds like hair-splitting. In practice it redraws who in the building touches what, and how much of anyone’s day vanishes into typing figures a machine could have read in a millisecond.
Where To Start
Don’t try to fix everything at once. Pick the single document type bleeding the most hours, usually supplier invoices or goods-received notes, and run only those through extraction to start. Clock how long reconciliation takes before, then after. If the number falls, you’ve got your case, and you widen the net from there.
The figure you needed was printed on that label before the pallet ever left the dock. Capturing it was never the hard part. Reaching it is the only step anyone keeps getting wrong.
