AI analytics for logistics companies: questions worth asking every week
By SchemaChat (Data BI LLC) · Published 31 October 2026 · Last reviewed 31 October 2026 · 8 min read
Logistics teams already hold the data in a transport management system or database export. The work is asking the same dozen questions every week and trusting the answers. This guide lists the KPIs with their formulas, twelve weekly questions, the data-quality traps that distort logistics exports, and how to check a number before you send it to a carrier.
We give no benchmark ranges here. Good values depend on lane, load type and service level, and we found no reliable public benchmark for small-carrier cost per mile. Use your own history as the baseline.
The KPIs and how they are calculated
| KPI | Formula | Watch out for |
|---|---|---|
| On-time delivery rate | Deliveries on or before the promised date ÷ deliveries with a delivery date × 100 | Missing delivery dates counted as late or on time |
| On-time in-full (OTIF) | Orders on time and complete ÷ orders delivered × 100 | Judging per line instead of per order |
| Cost per mile | Total transport cost ÷ miles driven | Fuel invoices paid in a different month from the miles |
| Capacity utilisation | Capacity used ÷ capacity available × 100 | Weeks sorted as text |
| Average delivery distance | Total distance ÷ number of deliveries | Cancelled deliveries included |
| Order cycle time | Delivery date − order date, averaged over delivered orders | Undelivered orders silently missing |
Twelve questions to ask your data every week
- What was our on-time delivery rate last week, by carrier?
- Which lanes had the most late shipments, and how many shipments are behind each percentage?
- How does cost per mile this month compare with last month?
- Which shipments have no delivery timestamp?
- Which customers' orders are late most often?
- What is the average and the 90th-percentile order cycle time by region?
- Which carriers changed the most in late deliveries this week?
- How did capacity utilisation change by week?
- Which products or customers most often cause short deliveries?
- What is freight cost per unit by carrier on the same lanes?
- Which suppliers deliver later than promised?
- How many shipments have duplicate rows in the export?
A worked example with a shipment table
Suppose your export has one row per shipment: shipment ID, carrier, origin, destination, promised date, delivered date, cost and miles. The question "which carriers were late most often last week?" needs only the rows with a delivery date, grouped by carrier, with counts next to percentages. A carrier that was late on 2 of 2 shipments is not worse than one late on 40 of 400, and a good answer shows both numbers.
For cost per mile, sum cost and miles per month first, then divide. Averaging the per-trip ratios weights a 5-mile trip the same as a 500-mile one.
Data-quality traps in logistics exports
- Duplicate shipment rows when a shipment has several stops or status updates.
- Missing or placeholder delivery timestamps.
- Promised dates overwritten when a delivery is rescheduled, which hides lateness.
- Time zones that move a delivery to the next day.
- Partial weeks compared with full weeks.
Check the number before you send it to a carrier
- Look at the query or calculation behind it and confirm the filters.
- Spot-check three shipments by hand against the carrier's own proof of delivery.
- Confirm the count of excluded rows (no delivery date, cancelled).
- Keep the calculation with the number so a dispute can be answered.
Limits
An analytics tool answers from the data you connect. It will not know where a truck is unless live position data is in your database, and it cannot explain a delay that nobody recorded.