One Line, Not One Order: How a Fulfilment Error Rate Is Counted

Two warehouses can both be honest about an error rate and produce numbers forty times apart, because the denominator is a choice. A provider quoting per-order accuracy is counting orders; a provider quoting per-line accuracy is counting the units inside them, including the three lines in a multi-item basket that most buyers never see. Our published figure is the stricter one, and the difference is worth understanding before it is used to compare two quotes.

Key facts

• The published figure is mis-picks below 0.01 percent, measured per order line. A line is one stock-keeping unit on one order, so an order of five units contributes five counted events rather than one.
• The response commitments are separate numbers. A reply on an inbound query inside two hours and resolution of an exception inside twenty-four hours are floor service levels, not accuracy statistics, and they are measured from the timestamp of the message rather than from the shipment.
• Discrepancies are reconciled on a fixed cadence, not on request. Stock differences are surfaced against the receiving record and the outbound record, and the reconciliation is a scheduled review of both rather than an argument in a chat thread.
• Recovery is bounded by the value declared at booking. Accuracy figures describe process; money follows the declared value on the shipment record and the terms page.

What one line means

An order line is a unit-level commitment: this stock-keeping unit, this quantity, this destination. When a picker scans a barcode at the pick face, the scan either matches that commitment or it does not, and that event is the atom of the counting model. A five-unit order that ships four correct items and one substitution produces one failed line and four clean lines — a basket that failed for the customer while contributing a single line to a rate measured across tens of thousands of them. The reason we publish the line-level figure is not that it flatters us: a multi-unit basket is where errors concentrate, so counting per line raises the number of chances to fail rather than lowering it.

The corollary: ask which denominator a provider uses before treating two percentages as comparable, since a per-order figure can sit on top of a per-line figure an order of magnitude worse.

Four error classes, four different fixes

A discrepancy between what was promised and what arrived is not one problem. The classes below are tracked separately in our exception record because each has a different root cause and a different owner, and because lumping them together is how a warehouse keeps fixing the wrong one.

The reconciliation that turns a complaint into a number

A dispute arrives as a claim about one order; what settles it is whether the records on either side agree. Our outbound record carries the pick scans, the carton weight and the label association with timestamps. The seller's side carries the store order, the channel feed and, often, a customer photograph. Reconciliation lines these up per unit and produces one of four outcomes: our error, a data error upstream in the feed, a carrier event after handover, or an unresolvable gap where the evidence does not reach.

Inventory discrepancies run on the same logic at a different cadence. A stock difference between the system and the shelf is reconciled against the receiving record and the outbound record for that unit; if the two do not explain it, a cycle count is raised and the adjustment is written with an audit trail rather than posted silently. A warehouse that always matches is usually one that adjusts without recording why — which is a reason to ask how an adjustment is logged, not whether it is.

What the numbers do not cover

An accuracy figure describes process on the floor, and stops at the floor's boundary. It does not describe carrier handling, a customs hold, a failed delivery attempt or goods defective before receipt. It is also not a compensation formula: the rate says how often a unit is wrong, while recovery follows the value declared at booking and the terms of that service. Sellers who read a process metric as an insurance limit are the ones surprised by a claim outcome.

Condition assessment is a second limit. We judge what we can see at receiving and at the pick face; a unit sealed in a supplier's retail box is not certified as working, and a fault found after a customer opens it is evidence about their experience, not automatically about the shelf.

Three questions about counting errors honestly

How do you define an order line, and why publish that instead of per-order accuracy?

A line is one stock-keeping unit on one order, which means a three-item order contributes three counted events. We publish the line-level figure because it is the harder one to keep low: an error that a per-order metric hides inside a mostly-correct basket becomes a visible failure when every unit is counted. It also matches how the floor works, since a pick scan resolves at unit level. The takeaway is not that our figure beats a competitor's, but that two percentages with different denominators cannot be compared at all — so ask the definition before the number, and ask it in writing.

My stock count in your system does not match my own spreadsheet. Which one is wrong?

Frequently neither, and the reason is timing rather than arithmetic. A store accepts an order on a channel feed, our floor releases it on a run, and the two systems stamp those events with their own clocks. A snapshot taken between the two disagrees with the warehouse by exactly the units in flight. The reconciliation that settles it is unit-level: the receiving record for the batch, the outbound scans since, and the adjustments logged in between, each with a timestamp and an audit trail. A discrepancy that survives that comparison is a genuine gap, and is treated as one — a cycle count against the shelf and an adjustment with a reason, not a quiet correction.

Does a low mis-pick rate mean you will pay for a bad shipment?

No. The rate describes how often the floor gets a unit wrong; recovery describes what happens when a shipment goes wrong, and those follow two separate documents. What is paid on a claim is bounded by the value declared when the parcel was created and by the liability terms of the service actually bought, which is why our terms page separates liability by lane rather than quoting an accuracy figure at you. If a seller wants recovery to track real transaction value, the lever is upstream — declared value set properly at booking — because it cannot be re-opened after a loss. The accuracy number is there to describe process discipline, and nothing more.

Read at 2026-10-30. The service levels here are the response and resolution commitments published on this site, and the accuracy figure is measured per order line; neither is a compensation term.

Published under SBT FulfillDock by Shenzhen Shibotong International Logistics Co., Ltd.

Related: how our three China warehouses are set up, what quality control covers, and the liability terms.