Most route operations measure too many things badly. A tight set of well-defined metrics, reviewed weekly by people who can act on them, beats a 40-tile dashboard nobody opens.
The twelve
Productivity
1. Stops per hour — completed stops divided by on-route hours. Segment by density band; a single fleet-wide number hides everything that matters.
2. Drive time percentage — driving as a share of total shift. Typical delivery operations run 35–55%. Rising drive percentage with flat volume means territory or assignment drift.
3. Cost per stop — fully loaded route cost divided by completed stops. The number to show finance, and the one to segment by window width, area type and customer segment.
Plan quality
4. Plan-versus-actual time variance — actual route duration minus planned, as a percentage. The single best health indicator of your data model. Persistent one-directional variance means your service times are wrong.
5. Planned-versus-actual distance variance — divergence here usually means drivers are not following sequences, or your road network data is stale.
6. Manual edit rate — percentage of stops moved by a planner after optimisation. Above roughly 10% and you have missing constraints, not a stubborn planner.
Service
7. On-time window compliance — deliveries completed inside the promised window. Report the miss distribution too, since a fleet at 94% with a 12-minute median miss is in a different position from one at 94% with a 90-minute miss.
8. First-time delivery rate — successful on first attempt. In consumer delivery this is often the highest-value metric in the whole set.
9. Failed delivery reasons — categorised and trended. Aggregating "failed" without a reason code makes the metric useless for action.
Utilisation
10. Vehicle fill — volume, weight or pallet utilisation against capacity. Watch both mean and distribution; a fleet averaging 70% with half the routes at 95% has a balance problem, not a capacity problem.
11. Route hour spread — the gap between longest and shortest route. The cleanest indicator of territory balance.
Cost
12. Overtime hours per route per week — the fastest-reacting cost signal in the operation, and usually the first thing to move when a plan degrades.
Definition traps
Metrics fail more often through definition than through measurement.
| Metric | The trap |
|---|---|
| Stops per hour | Including or excluding depot time changes the number by 10–20% |
| On-time | Arrival or completion? Grace period? Whose clock? |
| Cost per stop | Fully loaded or driver cost only? Are failures counted as stops? |
| First-time rate | Are customer-caused failures excluded? They should be reported separately, not removed |
| Drive percentage | Does breaks count as on-route time? |
Write the definitions down, publish them, and freeze them for at least a year. A metric whose definition drifts is worse than no metric, because it manufactures false trends.
Building a review that changes behaviour
Weekly, 30 minutes, same people. Operations manager, planners, a supervisor, and someone from customer service.
Four exhibits, in this order:
- Trend of plan-versus-actual variance, by depot
- Route hour spread, with the worst three routes named
- Failed deliveries by reason, week on week
- Overtime hours, with the routes driving it
One rule: every exhibit ends with an owner and a date, or it comes off the pack. Metrics without owners become wallpaper within two months.
Reporting maturity
| Stage | Looks like |
|---|---|
| 1 — Reactive | Numbers pulled manually when someone asks |
| 2 — Scheduled | Weekly pack produced, mostly descriptive |
| 3 — Diagnostic | Variance analysed to root cause, actions tracked |
| 4 — Predictive | Model decay detected before it hits service |
| 5 — Automated | Service times and constraints self-adjust from actuals, with review |
Most operations sit at 2 and believe they are at 3. The distinguishing question: when a route runs 40 minutes over, does anyone find out why, and does anything change?
What not to measure
- Vanity totals. Total kilometres saved since go-live is a press release, not a management metric.
- Driver league tables on speed. They incentivise exactly the behaviour your safety programme is trying to remove.
- Optimiser-reported savings. The engine comparing its plan to its own naive baseline is not evidence.
- Everything at once. A dashboard with 40 tiles is a way of avoiding the four numbers that matter.
Questions readers send us
How many KPIs should a route operation track? Six to twelve actively managed, with a deeper set available for diagnosis. If your weekly pack has more pages than the meeting has minutes, it is documentation rather than management.
What is a good stops-per-hour figure? There is no universal answer, and any vendor quoting one is guessing about your business. Density, drop size, access difficulty and service model swamp everything else. Build density bands from your own data and set targets per band.
Should drivers see their own metrics? Yes, provided the metrics are fair and the comparison is like-for-like. Route-level context matters: a driver on a dense urban round and one on a rural round are not comparable, and publishing them side by side destroys trust in the whole scheme.
How do I measure the value of the routing system itself? With a frozen pre-implementation baseline over at least four representative weeks, held to constant definitions. Retrospective baselines are always disputed and usually flattering. See building an ROI case.
What is the earliest warning that a route plan is degrading? Plan-versus-actual time variance, followed closely by overtime. Both move weeks before customer complaints or cost reports show anything.
