Delay detection starts with event definitions
AI should not classify delay abstractly. It should measure where the process paused:
- Vehicle docked but loading not started
- Loading started but no movement for threshold duration
- Loading complete but vehicle not departed
Timeline instrumentation
Each session is represented as a timeline with timestamps and contextual tags (dock, shift, carrier, workload type). This converts a subjective delay discussion into measurable operational evidence.
Rule strategy
Delay rules should be:
- Workflow-specific
- Threshold-driven
- Explainable to supervisors
Practical Operations Score set
- Average pre-load wait
- Active loading duration
- Post-load dwell
- Exception count per shift
Implementation note
Start with one dock cluster, tune thresholds for two weeks, and then scale once false positives are controlled.
