The customer ordered twelve cases of the 500ml. The van arrived with twelve of the 330ml — or ten of the right one, or the neighbouring route's cage entirely. Everyone involved is competent, the stock was in the building, and yet the doorstep got it wrong. Load errors feel random, which is why they get treated as inevitable; walk the goods from pick face to tailgate, though, and the errors cluster at a handful of predictable seams.
This article maps those seams, separates the two metrics that get blurred into one — picking accuracy and load accuracy — and sets out the verification gates that catch errors while they are still cheap. Because the same mistake costs pennies at the pick face, pounds at the tailgate, and a customer relationship at the door.
Two metrics, not one
Picking accuracy asks: did the right products, in the right quantities, in the right condition, come off the shelves for this order? Load accuracy asks a different question: did the right assembled orders get onto the right vehicle for the right route? Operations that track only a blended “delivery accuracy” number cannot tell a mispick from a mis-load — and the fixes live in different places, owned by different people, at different points in the morning.
The seams where it breaks
- Lookalike SKUs at the pick face. Same brand, adjacent variants, near-identical cases. Eyes tend to confirm what they expect to see; a scanned barcode is a far stronger system check on what is actually in hand.
- Units of measure. The order says 12 — twelve eaches or twelve cases? Products handled in two units at once are a standing ambiguity, and every ambiguous line is resolved by guesswork under time pressure.
- Batch and date rules applied by memory. Where stock rotation matters — first-expired-first-out for short-life goods — the “right” case is not just the right SKU but the right batch. Rotation enforced by habit produces correct picks most days and quiet exceptions the rest.
- Unrecorded substitutions. The picked item is out, so somebody sensibly substitutes — and tells no system. The delivery note, the invoice and the customer now hold three different beliefs about the order.
- The staging area shuffle. Assembled orders wait as anonymous-looking stacks. Two routes staged side by side, one interruption, and a cage crosses the line. This is the classic wrong-van error, and it happens after picking was done perfectly.
- Last-minute additions at the tailgate. The late order, the “stick these on for Thursday’s call” — goods that board the van through the side door of the process, on nobody's load list, in nobody's van stock.
The gates that hold the line
Gate 1 — verify the pick, item by item. A barcode scan at the pick face verifies the item in hand against the configured SKU, unit and — where enabled — batch rules, catching the lookalike and the wrong unit at the moment of picking; the control is only as good as the product data and validation rules behind it, which is an argument for maintaining them, not for skipping the scan. Crucially, it also gives substitutions a legitimate route: an out-of-stock becomes a recorded substitution instead of a silent one, so the delivery note and invoice stay truthful.
Gate 2 — verify the load against the route. Every assembled order scans onto a specific vehicle for a specific route. Scan the wrong cage and the mismatch is flagged before the load can be confirmed — the neighbouring route's goods no longer board silently. And the operating rule the gate exists to enforce: a late addition joins the load list — and the van's stock record — before it travels. This gate is what turns “the van’s contents” from folklore into a ledger.
Between the gates: keep the stages distinct. Pick, pack and load are three states, not one blurred activity. When an order's state is explicit — picked but not packed, packed but not loaded — the morning stops depending on the memory of whoever staged the cages, and a supervisor can see at a glance what is genuinely ready for the tailgate.
Two supporting habits multiply the gates' effect: load in reverse drop order where the load type allows — pallet and cage access, load stability, temperature zones and weight distribution all get a say — so the driver is not excavating at every stop (a dwell-time lever as much as an accuracy one), and blind-count the van periodically — a count where the expected figure is hidden — so van stock records are audited by evidence rather than confirmed by suggestion.
What an error costs at each stage
The economics of the gates rest on one asymmetry. A mispick caught at Gate 1 costs a walk back to the shelf. The same mispick caught at Gate 2 costs a repick under departure pressure. Caught at the doorstep, it costs a short or a refusal — a redelivery stop, a credit note, an invoice query and the payment delay behind it. Caught never, it becomes shrinkage: stock the system says exists and the shelf says does not, surfacing months later as a write-off nobody can explain. Every stage an error survives typically increases what it costs to fix, which is why the two gates are such high-leverage control points for whatever accuracy budget you have.
Where RouteMagic fits
RouteMagic's Warehouse App runs the flow exactly as described: pick, pack and load are distinct, barcode-first stages with their own states, so Gate 1 verification happens at the shelf and amendments are captured on the order rather than in the margins, and load orders tie each assembled order to its route and vehicle for Gate 2. Once configured per product, batch and expiry rules put FEFO enforcement into the scanner rather than the picker's memory — the discipline short-life stock depends on. What boards the van becomes the van's own stock ledger, reconciled at end of day and auditable by blind stock takes, with warehouse management reporting closing the loop back to error causes. Broader stock-control evidence from the same toolset — a whole-chain outcome rather than a Gate 1 / Gate 2 measurement, but the same disciplines at work: Bits 'N' Bobs reports eliminating 14% stock shrinkage after moving onto the platform (case study).
Conclusion
Load errors are not random; they are manufactured at six predictable seams between the pick face and the tailgate, and they obey a pricing pattern — every stage an error survives typically increases what it costs to fix. The response is not exhortation to be careful but architecture: split picking accuracy from load accuracy so you know which problem you have, put a scan gate at the shelf and another at the vehicle, keep pick, pack and load as explicit states, and give substitutions and late additions a recorded route into the process instead of a silent one. Start by measuring the two metrics separately for a fortnight and reason-coding every doorstep discrepancy back to its seam. The exercise often reveals that two or three seams account for most of your errors — and that the fix is narrower, and cheaper, than the folklore suggested.
