Enterprise motor carriers operating 250+ power units face relentless pressure to convert raw telematics streams into dispatch decisions and clean revenue recognition. Part 28 of the Applied AI for Enterprise Motor Carriers Systems Playbook translates three production-grade subsystems into concrete architectural patterns that integrate with Truckmate, McLeod, and TMW systems.
1. Real-Time Telematics Ingestion Pipeline
The foundation of any AI-driven operation is a fault-tolerant, low-latency ingestion layer capable of ingesting ELD pings, engine fault codes, and geofence events at scale. A modern carrier architecture routes Samsara, Geotab, and Motive data through a managed Kafka or Azure Event Hubs topic before landing in a staging schema.
| Column | Type | Purpose |
|---|---|---|
| event_id | uuid | Primary key |
| device_id | varchar(64) | ELD hardware identifier |
| truck_number | varchar(12) | Internal asset mapping |
| event_ts | timestamptz | UTC event timestamp |
| latitude | numeric(9,6) | GPS coordinate |
| longitude | numeric(9,6) | GPS coordinate |
| speed_mph | numeric(5,1) | Instantaneous speed |
| event_type | varchar(32) | PING, FAULT, GEOFENCE, HOS |
| payload | jsonb | Raw vendor payload for replay |
A materialized view refreshed every 60 seconds joins this stream to active loads (EDI 204) and driver settlement records, enabling downstream optimization engines to operate on fresh position data without repeated joins against the operational TMS.
2. Mixed-Integer Linear Programming (MILP) Dispatch Optimization
Static dispatch boards cannot account for live detention risk, fuel surcharge variability, or driver hours-of-service remaining. A production MILP model solved with Gurobi or OR-Tools replaces rule-of-thumb load-to-truck assignment with mathematically optimal decisions.
The objective function minimizes total cost = (linehaul + fuel_surcharge + detention_estimate) + (driver_settlement_delta) + (empty_mile_penalty) while respecting constraints on HOS, equipment type, and customer delivery windows. The model ingests the telematics view above plus carrier rate matrices stored in a normalized SQL schema.
Core Decision Variables
- xij = 1 if load i is assigned to driver j
- yjk = 1 if driver j deadheads to load k origin
- ti = projected arrival time at destination for load i
Constraints include EDI 214 status milestones, fuel tax IFTA state-mileage boundaries, and driver escrow balances. The resulting optimized plan is written back to the TMS via API or direct SQL update to the dispatch board tables in Truckmate or McLeod.
3. Straight-Through Billing and Revenue Recognition Pipeline
Once a load reaches EDI 214 Delivered status, the system must generate an accurate 210 invoice without manual intervention. The pipeline joins the optimized dispatch record, actual telematics-derived mileage, accessorial events (detention, lumper, tarp), and the carrier's current rate matrix.
Fuel surcharges are calculated using the exact OPIS or DOE rack average for the pickup date and the carrier's published table. Driver settlements are simultaneously generated using the same mileage and accessorial data, ensuring the P&L is clean at the load level.
Exceptions (revenue > $250 variance from original EDI 204, missing proof of delivery, or HOS violation) are routed to a managed exception queue rather than stopping the entire straight-through process. This design routinely achieves greater than 87% touchless billing for high-volume enterprise carriers.
Production Architecture Summary
The three subsystems form a closed-loop system: telematics ingestion feeds the MILP optimizer, which in turn drives dispatch decisions that generate clean billing and settlement records. All components are intentionally vendor-agnostic and can be layered onto existing Truckmate, McLeod, or TMW installations without replacing the core TMS.
Frequently Asked Questions
How long does it typically take to productionize a MILP dispatch optimizer for a 400-truck fleet?
With clean telematics ingestion already in place, the mathematical model, data pipeline, and TMS integration can be placed into production within 10 to 14 weeks. The majority of that time is spent validating business rules around driver settlement calculations and accessorial handling.
Can this architecture work with both McLeod and Truckmate simultaneously?
Yes. The ingestion and optimization layers are TMS-agnostic. Separate lightweight adapters translate the optimized dispatch plan and 210 invoice data into the specific database structures and EDI formats required by each system. Carriers running dual TMS platforms during mergers commonly use this pattern.
What data quality thresholds must be met before MILP optimization becomes reliable?
Geofence accuracy must exceed 94% and driver HOS events must be within 3 minutes of actual time. Fuel surcharge tables must be current within 48 hours. When these thresholds are met, the optimizer consistently produces lower empty-mile percentages than manual dispatch.
How does the exception management layer prevent straight-through billing from creating revenue leakage?
Any load that exceeds a configurable variance threshold (typically 4% or $250) is automatically removed from the automated 210 queue and routed to a revenue assurance analyst with all supporting telematics, rate matrix, and settlement calculations attached. The remaining loads continue without delay.
Strong Tower Consulting, LLC and Exspeedite LLC are separately organized sister companies under common ownership. Founded in Minnesota in 2008 by Scott Elliott, Strong Tower has spent more than 18 years helping enterprise motor carriers design and implement systems that deliver measurable operational leverage.
Our only product is better decisions.
Schedule a Systems AssessmentCall 763-244-1373 to speak with our team.
Continue the series: