Error Proofing in Warehouse Picking and Kitting

Picking errors split into two kinds that call for different answers. Wrong part — a similar-looking component in the same bin position, a substitute that was not authorised. And wrong quantity — three pieces instead of four, or a missing bag in a kit. Trying to solve both with one mechanism usually means solving neither well. Wrong part is fundamentally an identity problem and the cheapest answer is usually a barcode, unless packaging or label condition makes scanning unreliable. Wrong quantity is fundamentally a counting problem, and that is exactly where vision has the advantage. Separating the two before choosing the method is what makes the selection clear.

Barcode or vision for identity

Vision becomes the right answer for identity in a specific and common situation: labels that cannot be scanned reliably. Curved surfaces, reflective film, bags that fold over their own label, small components with no room for anything readable, or labels that a previous process has rendered illegible. In those cases the barcode step is not merely inconvenient, it is a source of errors, and vision judging shape, colour or markings can do the job directly. The limitation to state plainly is that vision struggles where two parts differ only in fine detail — a subtle difference in a moulding, a variation in shade between two similar materials — and those cases need either a barcode, a physical change to make the difference visible, or an explicit decision to accept the residual risk.

The resolution problem in kitting

Kitting verification is attractive because it looks like one picture: photograph the finished kit and check that every item is present and correct. The physical obstacle is resolution. A cart holding a dozen small parts may span a metre, and identifying parts a few centimetres across within that field needs a great many pixels; the pixel count on each part falls below what identification requires. Three practical responses exist. Photograph the kit in sections, accepting the extra capture time and the need to track which section is which. Constrain the layout with dividers or a labelled tray so every part has one fixed position, which turns a hard identification problem into a much easier presence problem. Or split the verification across the pick steps, checking each item as it is placed rather than the completed kit, which is often the most reliable and also catches the error at the point it happens. The trade-offs behind these choices sit in choosing a camera.

The design choices that decide whether it is used

• Time per pick: a verification step that adds noticeable time to every pick will be skipped as soon as the shift is running late. Aim for the time to be invisible rather than merely small.
• Feedback position: the prompt has to be where the operator is looking, not on a screen across the aisle. A signal light at the bin beats a message on a distant monitor.
• Cycle counting interruption: the opportunity to interrupt and correct has to be designed in. Whether the station lets the operator pause without losing the count, and how the system resumes afterwards, decides whether corrections actually happen.
• Error messages that are specific: "wrong part" leaves the operator guessing; "two short — bin B12" does not.
• A path for exceptions: urgent orders, partial picks, damaged stock. Without one, exceptions become a reason to bypass the system entirely, which then spreads.

Measure the baseline first

Before investing, measure the current error rate and, more importantly, where errors are found. That last detail usually changes the design. If most errors are discovered at final inspection, the recovery cost is high and step-level checking pays well. If they surface at the customer, the whole cost structure is different and the case is much stronger. Note also that some warehouses have low error rates already, simply because they are small or their products are easy to distinguish, and for them the honest answer may be that a verification system is not the best next investment. Saying so is more useful than selling one anyway, and the same accounting discipline appears in calculating the cost of an escape.