Autonomous Dispatch: Recommending the Right Driver for Every Load — in Seconds, With a Human in Command

Executive Summary

OnPace Automation partnered with a bulk cement carrier to eliminate one of dispatch's most time-consuming and error-prone tasks: figuring out, load by load, which driver is the right fit. In a dry-bulk operation, that decision isn't simple. The wrong driver can mean a contaminated load, a blown hours-of-service limit, an expired DOT inspection, or a costly deadhead — and dispatchers were carrying all of that in their heads, manually, for every assignment.

OnPace built a supervised autonomous dispatch engine — a recommendation system that scores and ranks eligible drivers for each load, then hands the final call to a human. It integrated directly into the client's existing TMS, so recommendations surfaced inside the tools dispatchers already used. Phase 1 was scoped deliberately to the bulk cement fleet, with an architecture built to expand without rework. Hard gates remove any driver who fails a compliance or safety check, weighted safety-first scoring ranks the remainder, and contextual modifiers tune the result for time of day, day of week, and home-time. The system recommends; the dispatcher confirms. The result: roughly 15 hours a week saved per dispatcher, more consistent and compliant assignments, and a full audit trail on every decision — with no loss of human control.

The Problem: A High-Stakes Matching Decision, Done Entirely by Hand

Assigning a driver to a bulk cement load is a multi-variable compliance and optimization problem. For each load, a dispatcher had to hold the entire picture in their head:

• Is the right equipment available? The correct tank and blower for the load.
• Is the last commodity compatible? White-load cement, grey cement, slag, and flyash can't be mixed — the previous load dictates what a truck can carry next without contamination.
• Is the driver eligible right now? Active status, and enough legal hours of service to complete the load.
• Is the paperwork current? Active, non-expired DOT inspections on both truck and trailer, and an active customer.
• And then the optimization: proximity and deadhead miles, gross margin, driver experience (especially on new customers), reload potential, plus time of day, day of week, and each driver's home-time profile.

Doing all of this manually, for every load, cost each dispatcher roughly 15 hours a week. It also meant decisions varied from person to person, and a single overlooked gate — an expired inspection, a contamination mismatch, insufficient hours — could turn into a compliance event or a ruined load.

The Vision: Recommend the Optimal, Compliant Driver — Instantly, Without Removing the Human

The goal was a system that could surface the best-fit, fully compliant driver for any load in seconds, while keeping a dispatcher firmly in control of the final decision.

That meant a tool designed around a few non-negotiable principles:

• Safety first — driver score and compliance dominate the scoring.
• Compliance before operations — compliance checks run before operational ones.
• Supervised autonomy — the system recommends, the dispatcher confirms.
• Pluggable inputs — built to absorb future data sources without a rebuild.
• Production-grade failure controls — treated as core requirements, not polish.

The Solution: A Supervised Recommendation Engine, Not a Black Box

OnPace Automation delivered a supervised autonomous dispatch engine, scoped in Phase 1 to the bulk cement fleet and integrated directly into the client's TMS. It runs as a supervised recommendation system: for every load, it evaluates the fleet, ranks the eligible drivers, and presents a recommendation — inside the dispatcher's existing workflow — for them to confirm.

The engine is built in layers:

• Hard gates — binary pass/fail filters that exclude ineligible drivers.
• Weighted scoring — a safety-first ranking of everyone who clears the gates.
• Contextual modifiers — real-time adjustments based on when the load runs.
• Operational reality inputs — the ground-truth data that feeds the gates and factors.
• Failure-mode controls — the guardrails that make it safe to run in production.

How It Works

Every load runs through the same pipeline, and a human confirms the outcome.

1. Hard gates (compliance first). Seven binary checks remove any unsuitable driver from the candidate pool, surfaced in priority order with compliance ahead of operations:

• Equipment match — the required tank/blower is available
• Commodity compatibility — the last load won't contaminate the next
• Driver status — active
• Hours of service — sufficient legal hours for the load
• Customer status — active
• Truck DOT inspection — current, not expired
• Trailer DOT inspection — current, not expired

2. Weighted scoring (safety-first). Every driver who clears the gates is ranked on weights that sum to 100%:

• 25% — Driver score (composite of compliance, safety, on-time delivery, refusal rate, reliability)
• 25% — Reload directional potential (single-next-load lookahead)
• 20% — Proximity to load (deadhead miles)
• 15% — Gross margin per load (a pluggable input, ready for future RAMS-adjusted margin)
• 15% — Driver experience (weighted higher when the customer is new)

3. Contextual modifiers. The scoring adapts to real-world timing (weights sum to 100%):

• 50% — Time of day (late-day shifts lean toward preserving hours of service and home-time)
• 30% — Day of week (Fridays weight heavily toward return-to-domicile)
• 20% — Driver domicile and home-time profile (home-daily vs. extended out-of-service)

4. Operational reality inputs. These carry no scoring weight but feed the gates and factors: plant/terminal operating hours at origin and destination, appointment vs. first-come-first-served handling, and a telematics health check (Samsara reporting freshness).

5. Confidence check and human confirmation. If the system's confidence falls below 80%, it escalates to a human dispatcher. Otherwise it presents its recommendation — and the dispatcher always makes the final call.

What Makes This System Different

This is supervised autonomy with production-grade discipline — not automation that removes the human.

• Compliance is gated, not scored. A driver who fails any safety or compliance check is excluded outright, before optimization even begins.
• The human stays in command. The system recommends; the dispatcher confirms. Manual override is always available — and every override is logged.
• Built for production failure modes. A confidence threshold, a full audit trail with input snapshots, stale-data detection on hours/location/load freshness, and a kill switch for full manual takeover.
• Explainable and accountable. Every autonomous decision is logged with the exact inputs that produced it.
• Future-proof by design. Phase 1 is deliberately constrained, with pluggable inputs (RAMS-ready margin) so later phases expand the system without a rebuild.

Measurable Results

• ~15 hours per week reclaimed per dispatcher, freed from manual driver-load analysis.
• Seven compliance and safety gates enforced automatically on every single load.
• Recommendations below 80% confidence auto-escalated to a human instead of being pushed through.
• A complete audit trail — every decision logged with its input snapshot.

Business Impact

• Dispatcher time redirected from repetitive analysis to exceptions, customer service, and edge cases.
• Lower compliance exposure. Hours-of-service, DOT inspection, and contamination risks are checked automatically before any driver is recommended.
• More consistent, explainable decisions, independent of which dispatcher is on shift.
• Scalable without headcount, on an architecture already designed to grow.

Built to Expand: Phase 2 and Beyond

Phase 1 was scoped tightly on purpose, with the next wave of capabilities recorded up front so they can be added without re-discovery — including DOT medical-card validation, CDL and tanker-endorsement checks, driver qualification file currency, insurance coverage validation, open critical DVIR defect checks, trailer-availability conflicts, and RAMS-adjusted margin scoring.

Why This Matters

For a bulk cement fleet, a mis-assigned load isn't just inefficient. A contamination mismatch can ruin product, an expired inspection or a blown hours-of-service limit is a compliance event, and a poor reload decision burns deadhead miles all week. Those judgments used to live entirely in a dispatcher's head, repeated for every load.

The system makes the safe, compliant, optimal choice the default — right inside the dispatcher's TMS, while keeping a human in command of every assignment.

The Takeaway

This isn't autopilot, and it isn't a black box. It's a supervised recommendation engine — one that enforces compliance, ranks the best-fit driver in seconds, and hands the final decision to a person.

The dispatch desk didn't lose control. It gained roughly 15 hours a week and a system that never skips a compliance check.