Mispicks are almost never carelessness. They are a slotting decision, a confirmation gap or a missing identifier — and all three are cheap to fix.
Key takeaways
Three changes, in this order, and none of them requires automation. Confirm at the face — the picker scans the location and the item before the system accepts a quantity. Slot against confusion — move look-alike items apart and put the fast movers in fixed, replenished pick faces. Make the identifier mandatory — the lot, batch or serial goes onto the pick record at the moment of picking, not later.
Everything else — pick-to-light, voice, robots — is a way of doing those three things faster. Do them slowly first and you will know whether the faster version is worth buying.
Mispicks feel like one problem and are usually four. Wrong item from the right location. Right item from the wrong location. Right item, wrong variant — size, flavour, voltage, lot. And right pick, wrong pack or wrong label. They have different fixes and the cheapest week you can spend is the one where every error is tagged into one of those four buckets before anything is bought.
This is a hazard-and-cause exercise and it benefits from being run like one. CCOHS describes job safety analysis as a process which helps assess a job to identify hazards and necessary controls, breaking the job into basic steps and identifying the potential hazards in each; the same discipline — break the job into steps, examine each — applied to picking rather than to injury will find the step where the error is being created.
Small warehouses usually slot for walking distance and then wonder why two similar items keep getting swapped. Distance is the second problem. The first is that a picker under time pressure resolves ambiguity by pattern, and two boxes of the same brand in adjacent faces are ambiguous.
The system-side tools for this are ordinary. You can divide your warehouse into zones to accommodate various storage needs, such as temperature requirements, or various turnover rates for items, and you can establish high-priority and low-priority storage areas for optimum placement of goods. Fast movers belong in fixed picking locations replenished from bulk — the standard pattern is that minimum/maximum replenishment is used for replenishment processes from a bulk area to the fixed picking locations.
A fixed face has a second benefit that matters for accuracy: the picker learns the location, so an unexpected item in a known face reads as wrong rather than as normal.
The cheapest slotting change
Separate the two items your team most often confuses. Not by a metre — by an aisle. Nothing else you do this quarter will have a better ratio of effort to error reduction, and it costs a lift truck and an afternoon.
Directed picking with confirmation is the change that moves the number. In a warehouse-managed operation, work templates define what work is performed, and how the work is done, and each must include at least one Pick operation and one Put operation, and location directives direct the work transactions to the appropriate location. The picker is told where to go and what to take, and proves both before the transaction closes.
Two design details decide whether confirmation actually catches errors. Scan the location as well as the item, or a picker standing in the wrong aisle with the right item still passes. And do not allow a manual override that is easier than the scan, because whatever is easier is what will happen at 16:45.
Handling units help too. Where the pallet or carton itself carries a label that the system tracks, the handling unit becomes the thing being confirmed, and a single scan carries the item, the quantity and the identifier together.
An accurate quantity with no lot code is only half a record, and in regulated categories it is a failure rather than a shortfall. Under the Safe Food for Canadian Regulations, documents must identify the food by common name, lot code or other unique identifier and trace it one step forward by indicating the date you provide it and the name and address of the person to whom you provide it.
The reason to capture it at the pick rather than at pack-out is the response window. On request those documents must be produced within 24 hours of the CFIA making the request, in English or French, and a lot code that was never recorded cannot be reconstructed from a shipped carton.
New and transferred staff are where the error rate lives, and orientation is a control rather than an onboarding courtesy. CCOHS treats worker orientation as the process of introducing new, inexperienced and transferred workers to their work areas and jobs — and a warehouse that hires seasonally is running that process several times a year whether it plans it or not.
The practical version: a documented pick procedure, a named person who checks the first hour of every new picker’s work, and a rule that a picker who finds a discrepancy stops and reports rather than adjusting silently.
Count errors per line picked, not per order — otherwise your accuracy improves every time an order gets bigger. Segment by picker, by zone, by shift and by item, because the aggregate number never tells you where to act. And measure the errors you catch internally separately from the ones the customer catches; the ratio between them is a better measure of your quality control than either number alone.
A New Brunswick 3PL with 18 pickers and no automation tags four weeks of errors. Fifty-eight percent are right-item-wrong-variant, concentrated on two customers with near-identical packaging. Twenty-two percent are right-item-wrong-location, all in one bulk aisle used as an overflow pick area.
Two changes follow. The confusable variants are moved to different aisles, and the overflow aisle is closed as a pick area in favour of fixed faces replenished on minimum/maximum replenishment. Location scanning is made mandatory, and the manual override is removed from the mobile menu. Lot capture is made a required field on the pick rather than a pack-out step, because the food customer’s obligation is to trace one step back and one step forward. No equipment was purchased.
The valuable work is diagnosis, not picking. Clustering the last few months of error records by item, location, picker, shift and hour so the concentration shows up without anyone building a pivot table. Proposing slotting changes for the item pairs that are actually being confused rather than the ones that look similar on paper. Reading a customer’s complaint email into a structured error record so the tagging happens at all. Drafting the customer response with the pick transaction and the lot record attached.
What it does not do is decide. Whether to move an item, whether a picker needs retraining, whether to accept a claim, and whether a lot should be held are decisions a person makes and signs.
Only if the location is scanned too. A confirmation that proves the item but not the place still passes a picker standing in the wrong face — which is why work is directed to a location by location directives rather than left to the picker.
Do the three changes manually. Fixed pick faces with a printed location label, a paper pick list that names the location as well as the item, a required lot field, and a second person checking high-value orders. The point of the software is consistency, not magic.
No — not without an authorisation step that is recorded. A substitution made silently is indistinguishable from a mispick when the customer receives it, and in a lot-controlled category it breaks the traceability chain the regulations assume, which requires the food to be identified by lot code or other unique identifier.
A 30-minute call is enough to tell you whether AI pays for itself in your back office.
