The news that moves freight
Tech & AI Nov 17, 2026 · 7 min read

Loadsmart puts numbers on freight AI agents

Per task, per load: the AI pricing model forwarders should copy

Loadsmart said on Monday 16 November that its gross profit grew 81% year on year in the third quarter of 2026. Over the same period its headcount grew 16%. That means gross profit per employee rose 56%. Gross profit, for a broker, is what is left after paying for the trucks and other services it buys on a customer’s behalf.

The Chicago freight technology company, which sells software and also runs a brokerage, credits AI agents. These are programs that carry out a task from start to finish, such as booking a dock slot or checking an invoice, rather than just suggesting an answer. Loadsmart says its agents handle about 171,000 tasks a month across its own freight operations.

What we think it means

Our view is that the most useful thing in the release is not the growth rate. It is the price tag.

Loadsmart sells its agents with “pricing per task or per load”. Most transport and forwarding software is still sold by the seat: a monthly fee for each person who logs in. That model makes less sense when the aim of the software is to need fewer people logged in. If an agent does the work of a coordinator, a per-seat price shrinks as the tool succeeds. Per-task or per-load pricing does the opposite. The vendor gets paid when freight moves, and the customer pays less when it doesn’t.

We think this is the model forwarders should demand from AI vendors, and copy when they sell digital tools to their own shippers.

The numbers behind the claim

All of the following figures come from Loadsmart’s release on PR Newswire, issued at 07:00 ET on 16 November.

  • Q3 2026 gross profit up 81% year on year.
  • Headcount up 16%.
  • Gross profit per employee up 56%.
  • About 171,000 agent tasks a month in Loadsmart’s own operations, covering 10,350 delivered loads in October.
  • 11 pre-built agents, across four stages of a shipment: “Plan & procure,” “Execute,” “Gate, yard and dock,” and “Settle.”
  • Agents put into production within 60 days.
  • A network of 1,800 customers, 4,500 warehouses, 230,000 carriers and more than 14 million dock appointments a year.

Divide 171,000 tasks by 10,350 loads and you get about 16.5 tasks per load. That is our arithmetic, not Loadsmart’s, and the two numbers may not cover exactly the same work. But it gives a feel for how many small jobs sit inside one truckload: tendering, tracking updates, appointment booking, paperwork, invoice checks. Forwarders will recognise the list. Each of those jobs is a few minutes of someone’s day.

The headcount figure matters as much as the growth figure. Gross profit per employee is the number forwarders and brokers live by. It is the clearest single test of whether technology is doing real work or just adding cost. In many forwarding businesses, profit tends to grow only when headcount grows with it, because more shipments mean more people handling bookings, documents and exceptions. A business where gross profit grows five times faster than headcount, as Loadsmart claims, is one where each person is handling far more freight than a year ago. If the claim holds, that is the gap that matters, not the 81% on its own.

Loadsmart says the same agents it runs internally are the ones it sells. Shippers can also build their own, and the agents can run on Loadsmart’s systems or on other vendors’ platforms.

Its forward claim is bold. Loadsmart “expects AI agents to handle more than half of that work inside an enterprise shipper’s TMS” within one to two years. A TMS is a transportation management system, the software shippers use to plan, book and pay for freight. Loadsmart adds that logistics teams would still set the rules and oversee the agents.

Felipe Capella, co-founder and CEO, put the pitch in one line: “Enterprise shippers want AI that comes with people who know freight,”

What the numbers do not show

The release gives growth rates but no revenue, no gross profit in dollars and no headcount total. So the 81% cannot be checked against anything, and a high growth rate on a small base can flatter. We also do not know how many of the 171,000 tasks would have needed a person, or how many were simple steps that older automation could already handle.

It is also worth remembering where Loadsmart has come from. In February 2022 it raised a $200 million Series D funding round led by SoftBank’s Latin America fund, at a $1.3 billion valuation, FreightWaves reported. That was close to the peak of the pandemic freight boom. A company with that valuation to defend has every reason to lead with efficiency numbers now. That does not make the numbers wrong. It does mean they are a sales message as well as a result.

The failure rate everyone is quoting

The release cites Gartner directly: “Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027.”

That forecast dates from June 2025. Reuters reported at the time, in a story carried by The Star, that Gartner blamed rising costs and unclear business value. Anushree Verma, a senior director analyst at Gartner, said: “Most agentic AI projects right now are early stage experiments or proofs of concept”. She also said: “Most agentic AI propositions lack significant value or return on investment”.

Gartner also warned about vendors that relabel chatbots and assistants as agents. It estimated that only about 130 of the thousands of vendors claiming agentic AI were genuine, according to the same report.

Loadsmart’s answer is to start small. It says it begins with narrow use cases, gets agents into production within 60 days and makes operators, not IT, own each project. Capella said: “Big visions lead to complex architectures that take years to build, and that’s where projects die,”

We agree with the diagnosis. In freight, the projects that stall are usually the ones trying to rebuild a whole booking flow at once. The ones that survive tend to take one painful, repeated job, such as dock appointments or invoice matching, and remove it.

What this means for forwarders

First, ask vendors for per-shipment pricing. If a software company says its agent will replace coordinator hours, it should be willing to be paid per task or per shipment. If it insists on per-seat fees, ask why the price does not fall as the tool takes over the work.

Second, measure your own AI the way Loadsmart does. Count tasks completed and shipments touched each month, and track gross profit per employee. Those three numbers will tell you more than any demo. A forwarder that cannot say how many tasks its quoting or booking bot finished last month does not yet know whether it works.

Third, expect shippers to ask the same of you. Large shippers will see releases like this one. Some will start asking forwarders what share of their own work is automated, and whether the savings show up in the rate. Forwarders that sell portals, quoting tools or visibility platforms should be ready to price them per shipment too.

Fourth, the target list is clear. Loadsmart’s four stages map closely onto a forwarder’s day: procurement and quoting, execution and tracking, terminal and warehouse appointments, and settlement. Settlement, meaning invoice checks, accessorial disputes and payment, is often the easiest place to start. It is rule-based, high volume and easy to measure. Dock and terminal appointments come next. Loadsmart handles more than 14 million dock appointments a year across its network, which shows how much of this work is repetitive booking rather than judgement.

Finally, be wary of the one-to-two-year claim for half of all TMS work. It comes from a vendor with a product to sell, and Gartner’s 40% cancellation forecast is a reminder that many projects never reach production.

Our prediction

We expect per-task or per-load pricing to spread. Our call: by 30 June 2027, at least two more freight software vendors will publicly price AI agents per task, per load or per shipment rather than per seat. We also expect at least one large forwarder or broker to report a gross-profit-per-employee or tasks-automated figure alongside AI claims in its results within the same period. If neither happens, the per-load model will have stayed a Loadsmart pitch rather than becoming the way freight AI is bought.

Sources

Read us on Google? Add Forwarder News as a preferred source to see more of our stories in Top Stories.