How predictions & risk work
Intelligence scores every tracked shipment as its events arrive. Two different questions drive everything you see. Keep them separate and the dashboard reads cleanly:
- Will it arrive by the promised date? This is the ETA and on-time prediction
(
p_late). It drives the map colors and the "delivers today" verdict. - Will it exceed the carrier's transit-time SLA? This is the SLA breach
prediction (
p_breachandsla_flag). It drives the Intervention Queue and the Act by deadline.
A shipment can be safe on one and at risk on the other.
The risk colors
The same language is used across every view:
| Color | Meaning |
|---|---|
| Green | On track or delivered |
| Amber | Watch, some risk building |
| Red | At risk: likely to miss, or an SLA flag, or already past its promise |
| Grey | Pending or pre-transit |
| Blue | In transit (on the map) |
The prediction panel
Open any shipment (from the Network Map) to see its full model output:
- ETA: a predicted delivery time with a confidence band (a low-to-high range), and a plain-language verdict for whether it makes today (YES, LIKELY, TOSS-UP, UNLIKELY, NO), with the probability behind it.
- Transit SLA: the breach flag,
P(breach), transit elapsed against the limit, and the last responsible moment (your Act by deadline). - No-scan risk: the chance the shipment is stuck (no new scan) over the next several hours.
- Misroute: whether the scan pattern suggests it has been mis-sorted, with a short explanation.
A WHY note explains the headline in words, for example "past its promised date with no delivery scan." The footer records when it was scored and the model version.
Rescore
Predictions update as events arrive, but you can force a fresh scoring run with Rescore on a shipment's detail panel. It is useful right after you know something changed.
Doing this over the API
Every prediction here is available programmatically:
GET /v1/shipment-predict?tracking=…returns the same model output.GET /v1/flagsreturns the Intervention Queue.GET /v1/overviewandGET /v1/carriersreturn the Overview KPIs.
See the API Reference.
Next steps
- Ingest & Scoring: how predictions get graded against reality.
- Work the Intervention Queue