In short
- For order-level picking accuracy, divide correctly picked orders by checked orders and multiply by 100. State whether you measured the first pick or the corrected contents, and publish the checked count alongside the percentage.
- Order, line and unit accuracy use different denominators and can all produce different results from the same mistakes.
- A mistake caught at the packing bench belongs in first-pick performance even when the customer receives the right goods.
- An unchecked order is unknown, rather than evidence of a correct pick.
- AirPost is a peer-to-peer fulfilment network: online shops have their orders packed and dropped at a parcel point by verified people near their customers, who earn a fee per parcel from home. No warehouse, no minimum order volume.
A percentage becomes useful when it tells you which part of the work needs attention. If a packer corrects the contents before dispatch, the customer outcome is good, but the original pick still required repair. If you mix those events, a busy bench can appear perfect while repeatedly fixing the same product confusion. This guide gives a small shop a defined measurement routine. The arithmetic examples are deliberately hypothetical; they do not describe AirPost customers or an industry benchmark.
Are you measuring orders, lines or individual units?
Choose the unit before creating the spreadsheet. An order passes only when every required pick within it is correct under your stated rule. A line measures a distinct order position, including the required variant and quantity. A unit measures the individual required pieces correctly fulfilled. These measures answer different questions: how many customer orders need repair, which positions cause problems and how much physical content is affected. Keep the definitions next to the result.
| Measure | Formula for the defined checked population | What one mistake can affect |
|---|---|---|
| Order accuracy | Correct orders ÷ checked orders × 100 | The whole order counts as incorrect |
| Line accuracy | Correct order lines ÷ checked order lines × 100 | Only the affected positions count as incorrect |
| Required-unit accuracy | Correctly fulfilled required units ÷ checked required units × 100 | The affected required pieces count as incorrect |
Do not interchange a unit pick and an order line. Several identical pieces can occupy one line, while two different variants occupy two lines. Also define how you record extra goods: an extra item is an order error even when all required units are present. A required-unit denominator alone would hide that overage. Keep an additional error flag for unrequested content and use it when classifying the whole order. Write the rule once, before reviewing results.
How can the same mistakes produce three percentages?
Consider a constructed counting example, not an observed shop workload. There are 100 checked orders containing 200 order lines and 500 required units. Two orders contain a wrong variant. Across those two orders, three lines and four required units are affected. Assume no extra items and no other faults. The same evidence now yields different percentages because each formula gives the error a different weight.
| Counting basis | Correct count | Arithmetic | Observed accuracy |
|---|---|---|---|
| Orders | 98 of 100 | 98 ÷ 100 × 100 | 98% |
| Lines | 197 of 200 | 197 ÷ 200 × 100 | 98.5% |
| Required units | 496 of 500 | 496 ÷ 500 × 100 | 99.2% |
If you report only 99.2%, someone may assume almost every customer order was flawless. In this example, two orders still needed attention. Use the order measure for that consequence and the other measures to diagnose the work. Compare like with like when reviewing a supplier or another period. A published percentage without its denominator, inspection coverage and error definition is insufficient for a meaningful operational comparison.
At what moment does the pick become correct or incorrect?
Set two checkpoints if your process includes a separate packing inspection. First-pick accuracy records the contents presented for that inspection before corrections. Release accuracy records the contents after any correction and before sealing. Link both observations to the same order reference. A correct release does not overwrite the first result. This distinction lets you see whether improved customer outcomes came from better picking or from more time spent repairing picks.
Keep customer-reported errors as a third event. A report received later can reveal a released error, but it can also need investigation. Record the allegation, the evidence and the confirmed classification separately. Do not count all returns as picking failures: a correctly selected item can be returned for another reason. Likewise, silence from a customer does not prove the original contents passed an inspection. Your report should describe what was actually checked.
Odoo 19 provides a concrete example of explicit criteria: a Pass - Fail quality check has instructions and a condition that must be met to pass. The operator selects Pass or Fail, and the workflow can create a quality alert after a failure. Apply the same clarity to your own form even if you use no quality software. “Looks fine” is weaker than a recorded comparison of article identity, required quantity and unrequested contents.
Does checking every order mean checking every item?
Not necessarily. Odoo distinguishes quality checks per Operation, Product and Quantity. Its example of one table and four chairs creates two product checks because there are two distinct products. Under Quantity, specifying 100% creates one check for the full quantity. The check count therefore does not automatically equal the number of individually inspected pieces. Describe inspection coverage separately from the number of records your software generated.
Odoo also offers All, Randomly, Periodically and On-Demand control frequencies. Those choices describe how checks are triggered, rather than a guarantee that all dispatched orders were inspected. In your report, show checked orders and eligible dispatched orders as separate counts. If you inspect only some orders, call the result sample accuracy. If you focus checks on difficult products, keep that targeted group separate from a randomly selected general sample.
An incomplete check should remain incomplete. Do not let a missing result become a default pass when exporting data. Note why the check was absent: interruption, unavailable evidence or a skipped step. These gaps are useful process information. They can tell you that the workload or handover is undermining the inspection routine, even before you can calculate a reliable error percentage. Fix coverage problems before celebrating a rising pass rate.
What does zero errors in a small sample actually tell you?
It tells you that you observed no errors in that sample. It does not establish a true error probability of zero. NIST explains that normal approximations can be inadequate when the sample or number of failures is small and describes exact binomial confidence limits. That matters for a small shop: an error-free short run can still be compatible with a noticeably larger underlying error rate.
For illustration, under a binomial model with independent observations and a stable error probability, a zero-error sample has a one-sided 95% upper error bound of 1 − 0.05^(1/n). Applying the binomial zero-error probability gives the calculations below. They were computed on 14 October 2026. These are one-sided confidence bounds for the error probability, not delivery guarantees, recommended error targets or two-sided intervals.
| Hypothetical independently checked orders | Observed errors | One-sided 95% upper error bound |
|---|---|---|
| 30 | 0 | 9.50% |
| 100 | 0 | 2.95% |
| 300 | 0 | 0.99% |
Use this as a reason to preserve the checked count, rather than adding confidence calculations to every daily task. Repeatedly scanning the same prepared order does not create independent evidence from new orders. A major change in products, staff or instructions also makes a historical combined rate harder to interpret. Show when the change occurred and keep the relevant periods visible. A statistic should help you investigate the process you have now.
Which error categories lead to a useful repair?
Record a short reason next to each failed first pick: wrong variant, wrong quantity, unrequested item or damaged stock selected. Keep a separate cause field that can remain unresolved until you investigate. The observed wrong variant is a finding; a confusing shelf label is a possible cause. Separating the two prevents a quick guess from becoming permanent evidence. Review the product, location and instruction associated with the event.
Change the part of the process supported by the evidence. Similar variants might need clearer separation or a better identity check. Repeated quantity errors might need an explicit unit definition. An incomplete handover might need an order reference that stays with the tray. Verify the revised routine with the affected order type and record when it started. Do not claim improvement merely because the next order happened to pass.
How do you put a released error into the cost calculation?
If a confirmed mistake requires one additional qualifying small parcel, the AirPost reference component is €2.25 recommended preparation plus €0.85 platform fee, giving €3.10 before its label and material. The constants were checked on 14 October 2026. This euro reference is not a GBP transport quote or a promised sender fee. Use the actual agreed fee and route when costing a real corrective dispatch.
Through AirPost for shops, you can agree storage, packing and handover with a sender. Agree which contents are checked and how an exception is reported; this article does not claim a built-in AirPost accuracy dashboard. Add actual replacement goods, customer support and any return costs only when you have their records. For charging units, read pick-and-pack cost per item or order. The error report and the invoice should refer to the same additional shipment.
Frequently asked questions
What is the order picking accuracy formula?
Correctly picked orders divided by checked orders, multiplied by 100. Define the checkpoint and error criteria. If you use lines or units instead, name that denominator explicitly.
What is a good picking accuracy KPI for a small shop?
Start with a defined measure and enough inspection coverage to interpret it. This guide does not invent a universal target. Report the checked count, first-pick failures and released failures together.
Do corrected mistakes count as picking errors?
They count as failures at the first-pick checkpoint. A successful correction can pass the release checkpoint. Keeping both results shows the work needed to prevent a wrong customer delivery.
Can I report 100% accuracy after an error-free sample?
You can report the observed sample result alongside its size and selection method. Avoid describing it as a guarantee about uninspected or future orders.
How do I improve order picking accuracy?
Investigate the recorded error type and verify a targeted change to location, identity, unit or handover instructions. Keep inspection coverage comparable so a smaller checked group does not create a misleading improvement.
Sources
Create your shop account and agree a specific contents check and exception report with a sender before handing over your first packing routine.
Agree your order check