The Friday operations report for a regional fulfillment team shows that 96 percent of orders shipped on time. A supervisor at the busiest warehouse has 89 percent on her board. Both numbers were pulled from working systems, and neither person has an obvious reason to distrust the other. Before asking the supervisor to explain a seven-point gap, the report owner needs to find out whether the two numbers describe the same work. If they don’t, any discussion about staffing or performance is starting with the wrong question.
This walkthrough uses a fictional two-warehouse network to show how to reconcile a discrepancy, document what changed, and decide whether the resulting number is ready for publication. The same approach works for help desks counting resolved tickets, field teams counting completed jobs, or finance operations counting invoices processed before a deadline. Start with one measure whose disagreement could change an actual decision. Reconcile that measure all the way back to individual records before expanding the exercise to the rest of the report.
Write a definition that two people can apply independently
Start with a sentence that answers four questions. What is the unit being counted? What makes it eligible? Which event determines the outcome? When does the reporting window close? For our network, the unit is an order and the eligible population is orders from both warehouses with a customer promise timestamp during the reporting week. Count an order as on time when its final required package gets its first carrier acceptance scan by that timestamp. The reporting week starts Monday at midnight and ends the following Monday at midnight, using the customer promise timezone as represented in the source system.
The last required package matters. If a customer orders a chair and a cushion and the items ship separately, counting a single early parcel would make the whole order appear on time. Likewise, printing a label is an internal milestone. A carrier scan is evidence that the package actually entered the carrier network. Site managers may still need the label timestamp to understand how the packing team performed. Give each measure a name that makes its event explicit and display the two rates separately if both affect decisions.
Write down exclusions as well. Our example excludes orders canceled before their promise timestamp because the customer was no longer waiting for shipment when the deadline arrived. It includes orders canceled after the deadline because removing them would erase an obligation the warehouse had already missed. If an order is reassigned between facilities, the report counts it once under the site responsible at the deadline. The reassignment rule determines whether a local manager can inspect the order and identify the team that owned the decision to transfer it.
Freeze the inputs before debating the outputs
Ask both owners for their source extracts with the reporting window, extraction time, and filters attached. Keep those extracts where an authorized reviewer can retrieve them. If one person reruns a live dashboard while the other keeps a Monday spreadsheet, you may introduce a new difference while investigating the old one. Record when each export ran and whether the source receives delayed events. An export at 8 a.m. might miss a Sunday carrier file that is usually delivered around noon.
For our example, request order ID, site, promise timestamp, cancellation timestamp, local label timestamp, and the carrier acceptance timestamp for the final required parcel. The extracts can omit customer names, addresses, and payment details. If the source tables contain several rows per package, flatten them in a documented step and retain the raw extract. Count the number of unique order IDs before joining anything. A jump from 600 unique orders to 612 joined rows can be a normal consequence of split shipments, but an order-level report must bring those rows back to 600 before calculating the rate.
Agree how to represent missing evidence. A blank scan could mean the package stayed on the dock, the carrier file failed to load, or the identifier was entered incorrectly. Put such records in a separate unknown category during reconciliation. Decide how that category enters the published figure and say so immediately beside the metric. In the fictional example, we count every eligible order in the denominator and show the number with unverified scans separately. A confirmed-on-time rate with unknown records in the denominator is a conservative estimate until those records can be resolved.
Build a bridge between the two totals
A reconciliation bridge changes one documented assumption at a time and shows the resulting numerator and denominator. Suppose Site East counted 534 timely orders out of 600 using label timestamps. Of the 600 orders, 30 were canceled before their deadlines, and 24 of those had timely labels. Applying the cancellation rule leaves 510 out of 570. Next, 18 of the 510 orders have carrier scans after the promise time despite early labels. Changing the event therefore leaves 492 out of 570, or 86.3 percent. Keep the order-level list behind each adjustment so the owner can review it without reconstructing the full query.
Then reconcile Site West using the same rules. Suppose 388 of 400 eligible orders there had timely carrier acceptance. Across both locations, the result is 880 confirmed timely orders out of 970 eligible orders, or 90.7 percent. A simple average of the two site percentages would give about 91.7 percent because it gives each warehouse equal weight. Use the pooled order count for the network rate and retain both site rates for accountability. A reader should be able to recompute the network rate directly from the counts printed in the report.
Watch for overlapping adjustments. If an order was canceled before its deadline and also had a late scan, changing both filters at once can make two people claim the same record explains their category. Apply the rules in the sequence published in the bridge. Once you have excluded an ineligible order, downstream changes to the scan outcome have no effect on the rate. When documenting an adjustment, include the new eligible total as well as the change in timely orders. That makes an overlap much easier to spot.
Decide how much weight to give the gap
The revised Site East number is lower than the local label-based figure. It doesn’t yet explain the underlying service problem. Pull a sample of orders that changed status and walk through the timeline with the warehouse lead. Check whether the parcels were already staged before cutoff, whether the scheduled pickup arrived on time, and whether the first scan took place at a sorting hub hours later. The operational response differs for an insufficient packing shift, a missed pickup, and a delayed scan feed. Record which explanation is supported by the evidence and who will investigate the remaining cases.
Suppose 12 of those 18 orders were on a trailer that left the warehouse on schedule, but the carrier’s file dropped the dock scan and picked them up downstream. Under the published definition, the orders are still late unless the agreed source supplies an earlier accepted event. Leave the definition intact for this period. Assign the carrier manager to investigate the missing events and document what supporting evidence would warrant a revision. That preserves the ability to compare periods while giving the team a way to dispute a source error.
Pick a trigger for withholding or labeling a figure before the next reporting cycle. In our example, the lead marks the total provisional when an expected carrier feed is missing or when unresolved orders could push the result across the threshold for a staffing review. If the trigger is 92 percent and even treating every unknown as on time would yield only 91 percent, the staffing review can proceed while the data issue stays open. If the possible outcomes fall on both sides, hold the staffing decision until the feed arrives or a reviewer documents an alternate basis for acting.
Publish the number with a traceable correction path
Place the final rate in the performance section of the Bit.ai Periodic Operational Report Template. Add eligible and timely counts for each site, the extraction timestamp, and a plain-language note about the difference between label creation and carrier acceptance. Link internally to the stored calculation and the approved definition. Under challenges, describe the specific source issue. Under next steps, name an owner and a date for the next check. A manager can read the summary in a minute, and a reviewer can follow the underlying records when a number is challenged.
For example, the published note could read, “Network on-time shipment was 90.7 percent, or 880 of 970 eligible orders. This week’s result uses carrier acceptance of the final required parcel. Site East’s local dispatch board uses label creation, so its rate measures an earlier step. Carrier Operations will review the 12 potentially missing dock scans and document the outcome by Wednesday.” If confirmed evidence changes the number, append a dated correction that states the old and new counts, the affected report period, and the reason. Tell the people who used the original result for a decision.
For the next cycle, have each site run a short prepublication check. Confirm the file arrived, apply the same period boundaries, count eligible order IDs, compare unknowns with the previous period, and reconcile totals back to the network number. Keep a brief record of which checks passed and who reviewed exceptions. A reporting owner who can reproduce last week’s rate should also be able to tell when a new disagreement comes from business performance rather than a quiet change in what gets counted. That is the difference between spending the weekly meeting debating the spreadsheet and using it to decide what needs attention.
Top comments (0)