Put these on your scorecard
- Order accuracy, inventory accuracy, and on-time dispatch are different measures.
- Suppose 12 shipments in a reviewed sample of 600 contain confirmed errors.
- Define the treatment of canceled orders, brand-requested holds, late stock, and incomplete addresses before calculating the score.

Start with the business question
Order accuracy, inventory accuracy, and on-time dispatch are different measures. Do not combine them into an attractive average before establishing what each measures. For order accuracy, decide whether an order fails when any item, quantity, packaging instruction, or destination is wrong. For dispatch, choose the event and deadline. A precise question prevents teams from optimizing a metric that does not represent the customer experience.
Expose the denominator
Suppose 12 shipments in a reviewed sample of 600 contain confirmed errors. The sample error rate is 2%; it does not establish the rate across unreviewed shipments. If only complaints were reviewed, the selection method is especially important. Publish the sample size and method beside the result. Do not quietly switch between orders, order lines, units, and parcels as the denominator changes from one month to another.
Handle exclusions consistently
Define the treatment of canceled orders, brand-requested holds, late stock, and incomplete addresses before calculating the score. Then report the excluded count. A warehouse team should be able to explain why an order left the eligible set. If a rule changes, annotate the date and avoid presenting the resulting improvement as though the process itself necessarily improved.
Use a score to trigger investigation
Break a headline metric into useful slices: facility, channel, product family, or issue type. Keep sample sizes visible so a small group does not look as conclusive as a large one. Choose a corrective action based on the pattern, then review the same measure after the change. Your scorecard should lead to a decision, not end with a green cell.
Record the result
- Write the numerator, denominator, time window, and exclusions.
- Keep order, unit, line, and parcel measures separate.
- Annotate definition changes before comparing trends.
A couple of questions
Can two 99% scores mean different things?
Yes. Different events, populations, and exclusions can produce identical percentages.
Should I average every KPI?
Only if the combined measure has a clear purpose and weighting; keep the underlying measures visible.
Reference and scope
Shopify: fulfilling orders — Background on fulfillment workflows and working with fulfillment services.
The recommendations are editorial guidance. Worked scenarios are illustrative, not reported provider results. How we research and handle featured placements.