Industries / Transport and vehicles / Trucking and freight / Freight brokerage and 3PLs
Matching loads and trucks
This is the work of finding a truck for a shipper's load, or a load for an empty truck, done by a firm that arranges the haul without running the trucks itself.
5
Milestones achieved, from 1985 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1985
A freight exchange on Minitel
Teleroute, then part of the French publisher Lamy, began listing loads and free trucks in France in 1985 on Minitel, the French national videotex network that ran on home terminals, under the service code 3614 Lamy. It later moved to computers and the web, and the same kind of open exchange spread across Europe. Teleroute is now part of the Alpega group, which says more than 350,000 users log into its software every day.
Credit
the Lamy group; Teleroute
2013
Korea limits resale of loads
Under a scheme South Korea introduced in 2013, a trucking firm that takes a load from a shipper can pass it on only once, must haul at least half of its yearly contracted freight itself, and, like licensed freight brokers, must report what it hauled and arranged to the transport ministry. Japan's transport ministry studied the scheme in 2025 as a model for its own rules.
Credit
South Korea's Ministry of Land, Infrastructure and Transport; described by Japan's Ministry of Land, Infrastructure, Transport and Tourism (MLIT)
2017
Software finds trucks to share
Tri-Vizor, a Belgian firm that sets up partnerships in which rival shippers combine their loads, uses a search tool that scans its database of more than 130,000 shipment lanes for loads that can share a truck, fill a return trip or make a round trip. Researchers Stefan Creemers, Gert Woumans, Robert Boute and Jeroen Beliën published the method in the journal Interfaces in 2017.
Credit
Tri-Vizor; Creemers, Woumans, Boute and Beliën
2024
Brokers change hands
UPS, which bought the broker Coyote Logistics for $1.8 billion in 2015, sold it to RXO for $1.025 billion on September 16, 2024, about 43% less than it paid. RXO says the deal made it the third-largest provider of brokered truck transport in North America. Earlier that year, in June 2024, the app-based broker Transfix sold its brokerage to the trucking company NFI and kept only its software business.
LessonA broker bought in a strong freight market can lose much of its value by the next downturn.
Credit
UPS; RXO; Transfix; NFI; reported by FreightWaves
2025
Japan records the real hauler
From April 1, 2025, Japan's amended trucking law requires a written contract that states the service and its price, including extra charges and fuel surcharges, and requires the firm that takes a load from a shipper to keep a register of the trucking companies that actually haul it. The rules also cover freight forwarders, Japan's licensed truckless operators. A second law, passed in June 2025, adds a duty to try to pass a load on no more than twice. In MLIT's 2023 survey of 4,401 trucking firms, more than 70% sometimes passed loads to other truckers, and about half of those hauling for a truckless forwarder were paid less than 90% of the price the first firm received.
Credit
MLIT; Japan's National Diet
Watch for
Pooling six shippers' loads
Japan's transport ministry took six months of delivery records from six shippers in different industries, all local deliveries around Tokyo, and calculated that planning their trucks together instead of separately would need about 13% to 14% fewer truck runs. At about ¥77,000 a run, it put the saving for the six at about ¥700 million in the slow season and about ¥900 million in the busy season. The calculation was done by the software firm Shimant and no trucks were actually pooled.
Achieved whenshippers combine loads on a shared system and an outside party reports the measured cut in truck runs.
Claimed by
MLIT study group on multi-tier subcontracting, June 2025
AI shippers crowd one carrier
In simulations, 50 shipper agents built on OpenAI, Anthropic and Google language models booked trucks for 30 days. On day one every model's first choice was the same carrier, which drew up to 76% of requests. Showing each carrier's remaining daily capacity cut this crowding by a third and doubled the shippers' gains.
Achieved whenanother group reproduces the result, or a freight platform publishes how AI-made carrier choices are spread.
Claimed by
Takahiro Ezaki, Naoto Imura and Katsuhiro Nishinari, July 2026
Open
01
A load-matching test for agents
An open set of matching problems, each a day of loads and trucks with the rules a dispatcher must keep (trailer type, weight, pickup and delivery times, driving hours left), each with the best plan worked out by an exact solver. An AI agent gets each problem through the tools a broker uses, such as a load list, a truck list and a booking call, and is scored on how far its plan is from the best one, how many rules it breaks, and whether it gets the same answer on 20 runs.
The briefCredit · unclaimed02
Rerun the pooling test openly
An open-source tool that takes several shippers' past delivery records in one shared format, plans their trucks once separately and once together, and reports the change in truck runs, empty running and truck hours. Any group of shippers could then test pooling on its own records without handing them to a platform.
The briefCredit · unclaimed
Open 01 · the brief
A load-matching test for agents
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of matching problems, each a day of loads and trucks with the rules a dispatcher must keep (trailer type, weight, pickup and delivery times, driving hours left), each with the best plan worked out by an exact solver. An AI agent gets each problem through the tools a broker uses, such as a load list, a truck list and a booking call, and is scored on how far its plan is from the best one, how many rules it breaks, and whether it gets the same answer on 20 runs.
Why it is still open
Published tests of AI agents in freight measure other things. A July 2026 simulation measured how agents crowd onto the same carriers, and a November 2026 tournament measured price negotiation; neither checks whether bookings obey the rules or how far they fall from the best plan. A GitHub search on November 20, 2026 found no freight-matching benchmark.
Built with what exists
Google OR-Tools or PyVRP for exact answers; USDA's free AgTransport lane data for realistic distances and rates; Python; small, low-cost models built for repeated tool calls (see the Note "The jobs agents repeat"); scoring that counts only problems passed on all 20 runs (see the Note "All twenty times").
First step
Write 20 problems of 10 loads and 5 trucks from one week of USDA lane data, solve them exactly, and run one agent through them.
Done looks like
At least 300 problems of 10 to 200 loads released openly, and a published table for at least three agents giving extra empty miles against the best plan, rule breaks per 100 bookings, and the share of problems solved correctly on all 20 runs.
Open 02 · the brief
Rerun the pooling test openly
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source tool that takes several shippers' past delivery records in one shared format, plans their trucks once separately and once together, and reports the change in truck runs, empty running and truck hours. Any group of shippers could then test pooling on its own records without handing them to a platform.
Why it is still open
Japan's transport ministry estimated 13% to 14% fewer truck runs for six shippers, but on private data and with a contractor's software. Tri-Vizor's partner-finding tool is its own and is not public. A web search and a GitHub search on November 20, 2026 found no open tool that does this.
Built with what exists
PyVRP or Google OR-Tools for planning routes; OpenStreetMap road data with the OSRM router for distances and times; Python; a generator for made-up delivery records.
First step
Write the record format from the measures in MLIT's report, generate one month of made-up deliveries for six shippers in one city, and plan them separately and together.
Done looks like
The tool and a made-up test set of six shippers over six months published openly; the separate-versus-pooled result on that set reproduced by a second person from the published code; and at least one real group of three or more shippers publishing its result with MLIT's measures (truck runs, share of time trucks are working, share of distance driven loaded).