Automate Broker Outreach, Keep the Dispatcher in Charge

Broker outreach looks like judgment work, but most of it is volume. A dispatcher covering a handful of trucks against a market of roughly 27,000 brokers spends most of the day typing the same email with different numbers in it, then chasing the brokers who did not reply. The typing scales badly. The judgment — which loads are worth chasing and what to say yes to — does not scale at all, and it should not.

The mistake teams make is trying to automate the wrong half. They either keep doing everything by hand, so outreach volume stays capped at one person's typing speed, or they hand the whole thing to a bot that emails brokers and accepts loads on its own. The useful answer is in between: automate the typing and the follow-up cadence, leave the decisions with the human. This post is about where that line sits and how to hold it.

What actually slows broker outreach down

Strip a day of outreach down to its parts and most of it is mechanical. The dispatcher scans load boards and broker inboxes, picks lanes that fit the fleet, opens a blank email, retypes the same intro with this load's origin, destination, equipment, and a rate, sends it, and writes the broker's name on a list to chase tomorrow. None of that is hard. It is just slow, and it competes for the same hours as the part that is actually hard: deciding which loads are worth the truck and which offers to take.

The volume math is unforgiving. With carriers averaging around $2.26 per mile in 2024 marginal cost (ATRI's 2025 report), every empty truck-day is expensive, and the way you fill trucks is by being in more conversations than you have time to start by hand. A dispatcher who can only send fifteen outreach emails before lunch is leaving the other forty viable lanes untouched, not because the loads are bad but because there were not enough hours to type. That is the constraint worth attacking, and it is a typing-and-tracking constraint, not a judgment one.

The follow-up problem is the same shape. Most broker replies do not come on the first email. They come on the second or third, sent at the right interval, after the load has aged a little and the broker is more flexible. Tracking who to nudge and when, across dozens of open threads, is pure bookkeeping that a tired human does badly. Drop a follow-up and you lose a load you already did the work to source. This is exactly the kind of cadence a machine keeps perfectly and a person does not.

Automate the typing and the cadence, not the decision

The clean way to think about it: the dispatcher decides, the system executes. Deciding means choosing which loads are worth pursuing, setting the floor on what you will accept, and approving each message before it leaves. Executing means surfacing the candidate loads, drafting the outreach at the dispatcher's direction, and running the follow-up sequence on schedule. The first column is judgment and stays human. The second is volume and can be automated without anyone losing sleep.

Here is the split made concrete.

The principle behind the table: automate the work that is repetitive and reversible, keep the work that is consequential and hard to undo. A drafted email that never sends costs nothing. A committed load costs a truck, a driver's day, and a broker's trust if you get it wrong. So the draft side can run wide open and the commitment side stays gated behind a person every single time.

What makes this safe rather than reckless is the approval step. The system can do everything up to the moment of sending, and then it stops and waits for the dispatcher. That one pause is the entire difference between an assistant that multiplies a good dispatcher's reach and a bot that emails brokers things the dispatcher would never have said. Numeo runs outreach this way on purpose: drafts and follow-ups are generated and queued, but they sit under the dispatcher's approval, not around it.

Pick the loads first, then let drafting scale

Outreach volume only helps if you are reaching out on the right loads, and choosing the right loads is judgment, so it comes first. Before any drafting happens, the dispatcher sets what the fleet wants: the lanes, the equipment, the rate floor, the brokers worth a relationship and the ones to skip. That is not a one-time setting you forget; it is the steering wheel. Markets move, the fleet's position moves, and the dispatcher adjusts what counts as a load worth chasing. The automation reads those preferences and surfaces candidates, but the dispatcher is the one saying yes, that lane, not that one.

This ordering matters because automating outreach on bad loads just produces more noise faster. The win is not "email more brokers." It is "email more brokers about the loads I actually want," which means the human's lane-and-rate judgment has to be in the loop before the drafting starts, not bolted on after. Get the targeting right and the volume engine amplifies good decisions. Get it wrong and it amplifies bad ones at the same speed, which is worse than doing nothing.

Once the targets are set, drafting is where the volume payoff lands. Instead of the dispatcher opening fifty blank emails, the system produces fifty drafts, each with the right load details and a rate anchored to the dispatcher's floor. The dispatcher's job shifts from typing to reviewing: glance, approve, edit, or kill. That is a far faster motion than composing from scratch, and it scales the reach of one person's judgment across many more conversations than they could ever start by hand. The judgment is still theirs on every message. They are just not retyping the boilerplate forty-nine times.

Let the machine keep the follow-up cadence

Follow-up is where automation quietly earns its keep, because it is the part humans drop. After the first round of outreach goes out, a chunk of brokers will not answer the first email. The discipline that wins those loads is a second and third touch at sensible intervals, and that discipline is exactly what falls apart when a dispatcher is busy. The load gets sourced, the first email gets sent, and then the thread dies because nobody remembered to nudge on day two. A sequencing engine does not forget. It sends the next touch on schedule, every time, across every open thread.

The boundary still holds here. The cadence is automated; the content and the limits are not. The dispatcher decides how many follow-ups a broker gets, how far apart, and when to stop pushing, because that judgment is about relationships and a machine should not be deciding how hard to lean on a broker you want to keep. The system runs the schedule the dispatcher set and surfaces the replies as they come, so the human spends their attention on live conversations instead of on remembering whose turn it is to be nudged.

There is a real reason to keep this on email rather than autonomous voice. With about 787,000 carriers in the market (FMCSA, December 2023) and 91.5% of them running ten trucks or fewer (ATA, 2025), broker relationships are personal and repeated. An email draft you approve before it sends is a message you can stand behind. An autonomous voice agent improvising on a call is a message you cannot take back and did not see coming. The whole model here is that nothing reaches a broker that the dispatcher did not approve, and email is the medium where that approval step fits naturally into the flow.

Why the boundary is the point, not a limitation

It would be easy to read "keep the human in the loop" as a hedge, a temporary state until the automation gets good enough to remove the person. It is not. The boundary is the design. Broker outreach sits on top of decisions, what a lane is worth, which relationship to protect, what offer to accept, and those decisions are the dispatcher's actual job. A median dispatcher earns around $46,860 a year (BLS, 2023) for exactly that judgment, not for typing speed. Automating the typing frees them to spend more of the day on the judgment, which is the opposite of replacing them.

This also lines up with the economics. Broker margins run around 13.5% on average (DAT, 2023), which means the room between a good outcome and a bad one on any single load is thin. A bot that commits to loads on its own will eventually take a bad one, and on thin margins a few bad commitments erase the gains from all the volume. Keeping a human on every commitment is not friction; it is the thing that protects the margin the volume was supposed to grow. The automation makes the dispatcher faster and wider-reaching. The dispatcher keeps the automation from doing something expensive.

So the takeaway is narrow and worth holding onto: automate the volume, never the judgment. Let the system surface the loads, draft the outreach, and run the follow-up cadence, so one dispatcher can be in many more conversations than they could type by hand. Keep the human deciding which loads to chase, setting the rate floor, and approving every send. That is the line, and the value comes from the system respecting it. This is exactly the boundary Numeo's AI Hub is built around: drafting and follow-up at scale, every commitment still the dispatcher's.