Client overview
- Industry focus
- Logistics
- Portfolio segment
- Logistics
- Organization profile
- Top-20 US 3PL, ~4k drivers, refrigerated and dry van
Shippers penalized dwell and late delivery with strict OTIF clauses. Dispatchers used three telematics portals side-by-side; planners lacked probabilistic ETA when ports and weather disrupted routes. Maintenance deferred decisions caused unplanned downtime spikes during peak produce season.
Problem
Telemetry fragmentation and naive ETAs drove dwell penalties and reactive dispatch — fuel burn exceeded benchmarks.
Drivers toggled apps that disagreed on available hours-of-service calculations. Geofencing accuracy varied by OEM device firmware, breaking automated arrival notifications.
Consolidated visibility for retail customers required manual analyst PowerPoints because API exports were stale by 30+ minutes.
Yard managers could not prioritize trailer doors when multiple high-priority loads queued; cross-dock staffing plans were disconnected from inbound certainty.
Solution
Canonical trip model ingesting telematics, TMS, weather, and port feeds; ETA service with calibrated uncertainty; dispatcher cockpit with automated playbooks.
Stream processing normalized events into a trip graph keyed by shipment ID with idempotent joins. ETA model blended graph routing, dwell priors per facility, and live traffic; confidence bands informed shipper communications automatically.
Driver experience unified in a Flutter app with offline-first task lists and DVIR flows synced when connectivity returned. Exceptions escalated via rules tied to SLA $ impact.
Fuel analytics correlated idle minutes with yard policies; recommendations surfaced to fleet managers weekly.
Implementation
- 1
Telemetry harmonization
Mapped vendor-specific event schemas to a canonical AVRO model; dead-letter queues captured malformed payloads without silent drops. Daily data quality dashboards flagged drifting odometers.
- 2
Pilot lanes & shipper sandbox
Two retail shipper lanes adopted shared ETA APIs with webhook subscriptions; sandbox mirrored production jitter for integration teams.
- 3
Operational adoption
Dispatcher coaching tied to fewer manual overrides; incentive alignment for maintenance to act on predictive alerts in 48h SLAs.
Tools & platforms
- Kafka
- Flink
- PostGIS
- Flutter
- Mapbox
- Snowflake exports for shippers
Engineering challenges addressed
- Handling GPS spoof attempts flagged by abrupt telematics discontinuities.
- Keeping ETA latency under 10s during nationwide weather incidents.
Program artifacts & environments
Tech stack
- Kotlin
- Flutter
- Kafka
- Flink
- PostgreSQL
- PostGIS
- Redis
- AWS
- Kubernetes
- Mapbox
Results
- 19% fleet-wide reduction in gallons per thousand miles within two quarters
- 27% fewer SLA penalties on monitored retail contracts
- Median dwell at top 20 facilities down 14 minutes
Quantified impact
19% fuel efficiency improvement
Normalized for freight mix via regression controls.
$2.4M annualized penalty avoidance
Attribution model from exception root-cause tagging.
Key takeaways
- Logistics software must reason about uncertainty — deterministic ETAs create false confidence for shippers.
- Driver UX is safety-critical; minimize context switching between compliance, nav, and messaging.
- Telemetry ingestion without observability becomes liability during disputes — invest in lineage early.