Restaurant Order Accuracy: Track Errors from Ticket to Handover

Choose the checkpoint and the cohort

An order can go wrong at the ticket, on the line, at packing or at handover. Pick one checkpoint and say so in the metric's name. Accuracy at handover to the guest or courier is a different measure from accuracy at the pass, and the two should not be combined.

Then define the cohort: which orders are eligible, for which channel and service period. Each order needs a stable ID from the order system so it can be counted once and traced back to its evidence. Agree what counts as an issue. A wrong item, a missing item, an unrequested change or an incorrect label are all examples. Write the definitions down before anyone starts recording.

Specify whether an issue corrected before release still counts as a detected issue at the checkpoint. Keep detected-and-corrected issues separate from errors that reached the guest if both are tracked. In the illustration below, an issue detected during the defined handover check counts even when corrected before release; the 94% figure is not a guest-visible error-free rate.

Orders and events are different counts

One order can have more than one problem. If you count issues and call the result the share of orders affected, you overstate it. Keep two counts: the number of distinct orders with at least one recorded issue, and the number of issue events.

The issue log records events, so it is not automatically a count of distinct orders. Reconcile the log against the stable order IDs before entering the summary. In the worksheet, the summary rows are manual totals from that reconciled record; the sheet does not deduplicate orders for you.

Give each distinct issue an event ID linked to its order ID. A later comment or a second observer recording the same missing item is additional evidence for that event, not automatically another event. Keep the issue count tied to the agreed definition, and restrict customer identifiers to the appropriate system rather than copying them into the shared worksheet.

Worked example: 50 orders at handover

In this fictional sample, 50 eligible completed orders were all observed at a defined handover checkpoint. Of these, 47 had no recorded issue and 3 distinct orders were affected. No outcomes were unknown. One affected order had two separate issues, so there were 4 issue events in total.

That gives 3 of 50, or 6%, of orders affected and 47 of 50, or 94%, clear at this checkpoint. The event rate is 4 divided by 50, which is 0.08 events per order. It would be wrong to report 8% of orders affected, because that divides events by orders as if each event were a separate order.

These figures describe only what was observed at that checkpoint. They say nothing about how guests experienced the meal, whether problems arose after handover, or whether the kitchen is safe.

Unknown outcomes and empty cohorts

Sometimes an order leaves without being checked. An unknown outcome is neither clear nor affected. If any orders in the cohort are unknown, withhold the full-cohort rates and report how much of the cohort was actually observed. Treating unchecked orders as clear inflates accuracy; treating them as affected deflates it.

If the cohort is zero, there is no rate. The counts for clear, affected and unknown orders must be whole numbers that add up to the cohort.

Safety does not wait for the weekly figure

An accuracy percentage is a review measure. Any issue involving an allergy or other safety concern follows the site's existing controls immediately, at the time it is found. Record it through the required incident process; add only the appropriate reference to the shared review log. Never delay the immediate response for a review meeting.

Turn findings into a targeted review

Look at the events by type, stage and order context rather than at the headline figure alone. If the same kind of issue recurs, such as a modifier missed on delivery orders, investigate that pattern with the people who do the work. Check the evidence before naming a cause.

Agree one corrective action, an owner and a date to recheck. After the change, compare a similar cohort at the same checkpoint. A lower figure from a quieter week, or from a period when fewer orders were observed, does not show that the change worked.

Where the method comes from

Quality guidance on check sheets describes structured data collection: decide which event to observe and define it clearly, decide when and for how long to collect data, design a simple labelled form, try it briefly to make sure it works, and record each occurrence as it happens. The order IDs, cohort and event definitions, and example in this guide are original applications of that general approach. They are not a restaurant metric standard or an accuracy benchmark.