AI Dispatch: Better Load Selection Without More Headcount

Most carriers think they have a load-volume problem when they actually have a load-selection problem. A dispatcher can only open, read, and price so many postings in a shift, so the team books from the loads it happened to look at — not the best loads available on the lane. The fix that gets reached for first is another dispatcher. That is the expensive way to widen the funnel, and it does not reliably raise the quality of what gets booked.

This piece is about the other path: improving selection quality by evaluating far more options against consistent rules, faster than a person can, so the same team books better freight. The dispatcher stays in control of every commitment. What changes is how many loads get a fair, rules-based look before one is chosen.

Selection quality is a search problem, not a volume problem

Booking more loads and booking better loads are different goals, and chasing the first one often hurts the second. When a dispatcher is measured on keeping trucks moving, the pressure is to take the first acceptable load, not the best one in the set. "Acceptable" usually means it clears a rough rate-per-mile floor and the timing works. That is a low bar, and it ignores most of what actually determines whether a load is good: the all-in rate after deadhead, where the truck ends up for its next load, and whether the broker pays on time and without drama.

The real constraint is attention. There are far more loads posted on a given lane than any one person can evaluate properly. So the dispatcher samples — opens a handful, prices a couple, books one — and the rest never get looked at. The booked load is the best of a small, arbitrary sample, not the best of what was available. Two dispatchers with the same trucks and the same boards will book very different freight depending on which postings they happened to open first.

Framed that way, load selection is a search-and-ranking problem. The quality of the decision depends on how much of the available set you can evaluate and how consistently you score each option. A human evaluates a small slice with judgment that drifts over a long day. Software can evaluate the whole slice against the same rules every time. That is the leverage — not replacing the dispatcher's judgment, but making sure the judgment gets applied to the best candidates instead of a random few.

What "good" actually means, made explicit

Most carriers carry their selection criteria in someone's head. The experienced dispatcher "knows" which brokers to avoid and roughly what a lane should pay, but that knowledge is informal, uneven, and walks out the door when they do. The first real step toward better selection is writing the criteria down as rules a system can apply to every load the same way.

A usable rule set goes well past rate-per-mile. The dimensions that separate a good load from a mediocre one on the same lane are concrete:

• All-in rate, not line-haul. The number that matters is total revenue minus the cost to run the miles, including the empty miles to pickup. A higher gross rate with a long deadhead can net less than a lower rate with the truck already in position.
• Deadhead to pickup. Empty miles are pure cost. The ATRI 2025 report put the marginal cost of operating a truck at roughly $2.26 per mile for 2024; deadhead commonly runs 15 to 30 percent of total miles. Both ends of that range are expensive, and a load that looks strong on line-haul can be average once the empty miles are counted.
• Where the truck ends up. A load into a thin market can cost you on the next booking. Selection quality compounds across loads, so the destination's outbound options belong in the score.
• Broker quality. A broker that detains drivers or pays slowly carries real cost. Detention alone runs an estimated $1.1 to $1.3 billion a year industry-wide. A broker score built from your own payment and dwell history belongs in the ranking.

Write those down and "good" stops being a feeling. It becomes a consistent function you can apply to every load on the board — which is exactly what makes the next step possible.

How automation widens the funnel without widening the team

Once the criteria are explicit, the bottleneck is throughput: applying them to every available load fast enough to matter. This is the part that does not scale with a single dispatcher, and it is the part software handles well. A system can pull postings from multiple boards, normalize the inconsistent fields — one broker leads with the rate, another with the appointment window, a third omits the load number — and score each one against your rules in the time it takes a person to open two tabs.

The output is not "here are 400 loads." It is a ranked shortlist: the loads that clear your rules, ordered by all-in value after deadhead, with the weak brokers and dead-end destinations already pushed down. The dispatcher opens the shortlist instead of the firehose. They are still choosing, still pricing, still deciding — but they are choosing from the best candidates on the lane rather than the first few that loaded. That is how the same headcount books higher-quality freight: the human's attention lands where it is worth the most.

This is also where Numeo keeps the line clear about control. The AI Hub ranks loads against the carrier's own rules, adds market context, and can draft broker negotiation messages — but the dispatcher approves the rate and the booking. The negotiation Numeo runs today is by email and message, not autonomous voice calls to brokers; a person sets the targets and signs off on commitments. The software widens and orders the funnel. The dispatcher still decides what gets booked, at what price, with whom.

The economics: a worked comparison

The instinct to hire is worth pricing out, because the numbers favor the other path by a wide margin. BLS 2023 put dispatcher pay at a median near $46,860 a year, about $22.53 an hour — and that is base pay before payroll taxes, benefits, software seats, and the weeks of ramp before a new hire knows your lanes and brokers. Call it well over $50,000 fully loaded in year one.

Here is what each approach actually changes:

The hire widens the funnel by adding another pair of eyes that still samples a slice of the board with criteria in their head. Rules-based selection widens it by evaluating the whole slice consistently and handing the dispatcher a ranked shortlist. The first scales linearly with payroll; the second scales with compute. For the typical small carrier — and 91.5 percent of US carriers run ten trucks or fewer, per ATA 2025 — adding a salaried head to fix selection quality is the costly tool for the job. The cheaper and more durable fix is to make the existing dispatcher's attention land on better loads.

There is a quality dividend on top of the cost difference. A consistent rule set does not have a bad afternoon, does not forget that a broker shorted you last month, and does not skip the deadhead math when it is busy. It applies the same standard to load number one and load number four hundred. That consistency is where the selection-quality gains come from, and it is the thing a second hire cannot reliably deliver.

Where the dispatcher has to stay in the loop

Better tooling does not move the commitment decision off the human, and pretending otherwise is how carriers get burned. Freight has too many context-dependent calls for a rule set to own the final say. The rules are very good at the part they are built for — scoring and ranking the available loads — and weak at the parts that need judgment the carrier cannot fully encode.

A broker relationship worth protecting can justify a load the rules would rank lower. A driver's home-time, hours, or a quietly preferred lane can override the all-in number. And freight fraud means the booking step itself needs human eyes: cargo theft hit an estimated $725 million in 2025 per CargoNet, with double-brokering on the rise, so verifying who you are actually dealing with stays a human responsibility no ranking score replaces. The rules narrow four hundred loads to a strong ten; the dispatcher picks among the ten, prices the load, vets the broker, and commits. That division of labor is the point — not automating the decision, but making sure the decision is made over better options.

The takeaway

If selection quality is the problem, more headcount is an indirect and expensive fix. The direct fix is to write down what makes a load good — all-in rate, deadhead, destination, broker quality — and apply those rules to every available load instead of the handful a person can open by hand. That widens the funnel and raises the quality of what gets booked without changing the size of the team. The dispatcher keeps every commitment decision; what changes is that those decisions are made over the best loads on the lane, ranked consistently, every time. That is the shift the AI Hub is built to make on the carrier's own rules.