Definition
Average total downstream delay minutes caused by mid-route AI re-routing decisions, netting the delay avoided by the re-route against the cascade delays imposed on all subsequent stops.
The Measurement Gap
Re-routing decisions are evaluated at the point of the re-route without tracing consequences to downstream stops. The blind metrics include:
Re-route success rate: measures whether the re-routed segment itself performed as predicted (avoided the original delay) but never follows the load through subsequent stops to detect cascade effects. A re-route that avoids a 30-minute traffic delay but causes the driver to miss a dock appointment window (adding 4 hours of wait) is counted as successful.
Average delay per load: captures total load delay as a single number but cannot attribute how much delay originated from mid-route re-routing versus other causes. The re-route contribution is invisible inside the aggregate.
Route adherence percentage: tracks how closely drivers follow planned routes but treats all deviations equally. AI-recommended re-routes and driver-initiated detours are indistinguishable, and neither is connected to downstream timing outcomes.
None of these metrics model the cascade: when a re-route changes arrival timing at one stop, every subsequent stop's planned arrival shifts. Dock appointment windows become misaligned, layover timing changes, and the cumulative downstream delay can exceed the delay the re-route was designed to avoid.
Decision Supported:
When AI re-routes a driver mid-trip, does the re-route save more time than it costs at downstream stops across the full remaining delivery sequence?
Formula
Plain Language
RCDI calculates the net time impact of each mid-route AI re-routing decision by summing the delays at all stops after the re-route point and subtracting the delay the re-route was estimated to avoid. A positive RCDI means re-routes are net-negative: they cause more downstream delay than they prevent. A negative RCDI means re-routes are net-positive: the avoided delay exceeds the cascade impact.
Notation
$$RCDI = \frac{1}{N} \sum_{i=1}^{N} \left( TotalDownstreamDelay_i - AvoidedDelay_i \right)$$
Where:
$$TotalDownstreamDelay_i = \sum_{s \in S_{after}} \left( actual_arrival_s - planned_arrival_s \right)$$
$S_{after}$ = set of all stops after the re-route point for load $i$
$N$ = number of re-route events in the measurement period
Variables
| Variable | Description | Source |
|---|---|---|
| TotalDownstreamDelay | Sum of arrival delays (actual minus planned) at all stops after the re-route point | Telematics/ELD geofence timestamps vs. TMS planned arrivals |
| AvoidedDelay | Estimated delay the re-route was designed to prevent at the immediate next segment | Routing engine logs (estimated counterfactual) |
| actual_arrival | Timestamp when the vehicle arrived at each stop (geofence trigger) | Telematics / ELD |
| planned_arrival | Originally scheduled arrival time at each stop before the re-route event | TMS planned schedule |
| N | Count of mid-route AI re-routing events in the period | Routing engine / TMS event logs |
| re-route point | The geographic and temporal point at which the AI re-route decision was executed | Routing engine event log |
Proposed formula, pending validation pilots.