⌂ / Outbound / Packing Bench

Packing Bench — Shipment PS-0831

✦ AI Prototype ← All prototypes
Proposal P2 · QC Photo Check — proven before promised. The packing bench is the one step where eHub itself enforces one scan per unit. Here, a single photograph counts the pieces and reads the printed labels — in shadow beside today's process. The AI never passes or blocks a shipment.
Shadow mode This check reads the same data the bench screen already shows and writes nothing. A match suggests a fast pass; a mismatch means recount — exactly what happens today. Nothing is promised before the week-one method study measures real accuracy on SNT's own products.

Shipment

today's bench data · read-only
PS-0831
SO/2609/000584
MONTIGO
3 · 63 pieces expected

✦ Photo verification — line by line

assist, never the judge
A-1042 Vitamin C 1000mg effervescent · 24 expected in single-layer tray Expected 24 Awaiting photo
Fixed tablet stand · single-layer layout

What the AI sees

Counting pieces and reading printed labels…
B-2210 Paracetamol 500mg 24s · promotion line Expected 36 Awaiting photo
Fixed tablet stand · single-layer layout

What the AI sees

Counting pieces and reading printed labels…
C-0871 Herbal supplement gift pack · serial-tracked Expected 3 Scanner — as today
▣Separated by design: serial-tracked products are permanently outside photo counting — every piece is scanned individually, exactly as today. The bench data already flags them line by line, so they are separated automatically, not discovered late.
Pilot agreement rate (shadow vs bench outcome): — would land beside the throughput figures supervisors already read — the first accuracy number the bench has ever had.

Week one is the method study — AI competes, it isn't crowned

honest comparison
MethodCountingReads labels?Setup per productWhere it fits
Counting scaleExcellent for uniform itemsNoUnit weight per productQuantity-only checks on consistent stock
Batch barcode imagerExact — for visible barcodesIdentity via barcodeNoneLabelled cartons, every label facing up
✦ AI photo assist (this proposal)Weakest pure counter herePrinted text — not barcodesNone — handles products it has never seenProduct variety — and the only method that also reads, compares and explains

Measured head-to-head per product category on real bench products, against targets agreed before the study starts. Whichever method wins a category is what we recommend for it — not everything needs AI. If nothing passes, the scorecard is the deliverable and we stop.

Prepared by Anchor Sprint Sdn Bhd · Proposal P2 · Draft v0.4 Interactive prototype · illustrative data only · not connected to any system