At four in the morning, a driver parked at a service area on the G60 refreshes a screen. Thirty tonnes of steel coil need to move from a yard outside Hangzhou to a factory in Jiangxi by evening. He is one of several hundred drivers looking at the same listing. A few seconds later the load is gone, claimed by someone whose empty run to the pickup point was shorter than his.
That invisible sorting — which truck, which load, in what order — is the product. Everything else about China's largest digital freight platform is bookkeeping around it.
Key takeaways
- Full Truck Alliance (满帮集团, NYSE: YMM) reported 2025 revenue of 12.49 billion yuan (≈ US$1.76B / HK$13.7B), up 11.1 percent, with net income of 4.46 billion yuan (≈ US$628M / HK$4.91B).
- Its in-house freight model, 满运大模型, now powers a voice-based "find a truck" service; the company says shipper–truck matching efficiency rose about 40 percent and average empty distance per order fell 27 percent.
- The number that best describes the business is not either of those: it is a 42.7 percent fulfilment rate, meaning most posted loads still do not get matched.
What the platform actually replaces
Before apps, China's road freight ran through 信息部 — small brokers with a whiteboard and a phone. A shipper called around, a driver drove to a market and looked at a board, and the price depended on who you knew.
Full Truck Alliance stitched two apps — 运满满 for shippers, 货车帮 for drivers — into one network, then spent years converting a listings business into a transaction business: electronic waybills, standardised contracts, escrow payments, and a commission charged once a deal closes.
That conversion is now the growth engine. Transaction service revenue reached 5.32 billion yuan (≈ US$749M / HK$5.85B) in 2025, up 38.2 percent, while commissions were live in 273 cities. In the fourth quarter, 63.9 million orders were fulfilled on the platform, 12.3 percent more than a year earlier, and average monetisation per order was 26.3 yuan (≈ US$3.70 / HK$28.9).
Why dispatch is a model problem, not a search problem
The hard part is not listing a load. It is ranking a load for a specific driver, given where he is, where he wants to end up, what he is carrying, what the traffic and tolls look like, and whether the shipper has a history of paying late or overloading.
Each of those signals used to be a rule or a human judgement. The company's pitch is that 满运大模型, built on its own transaction data, weighs them together: it ranks loads, predicts which will actually be fulfilled, and nudges both sides toward those.
The visible change came in the fourth quarter of 2025, when Full Truck Alliance piloted an AI assistant for shippers that lets them post a load by voice and have the platform dispatch it. The company later described a second layer at a cloud computing conference in September 2026: production agents that act as a driver's assistant, watching the market around the clock, and as a shipper's assistant that follows a shipment from posting to delivery. The unit of work, its engineers said, is "one commission" — the agent owns a task end to end rather than answering a question.
That framing is more revealing than the efficiency numbers: the model is there not to answer questions but to take over the loop a broker used to run — find, negotiate, confirm, track, settle.
The 42.7 percent problem
Every platform story needs a denominator, and this one is unusually honest. Fulfilment rate — the share of posted orders that actually get matched and delivered — reached 42.7 percent in the fourth quarter of 2025, up more than five percentage points year on year and a record for the company.
Read the other way, most posted loads still do not result in a truck. Some are duplicates, some are price tests, some are shippers who post to five channels and take the first yes. But the gap is the size of the opportunity: every point of fulfilment is revenue the platform does not have to acquire twice.
The same is true of empty running. The company says its "three empties" ratio — empty driving, empty positioning, empty return — fell from 38.97 percent in 2020 to 34.88 percent in 2025. That is real progress and still a third of the fleet's motion spent going nowhere useful. A 27 percent cut in empty distance per order, if it holds, is the kind of number that compounds: it is fuel, tolls and driver hours, not a software feature.
Money, credit and the limits of a good model
The 2025 results show a platform that has learned to make money after the listing. Revenue up 11.1 percent, net income up 42.8 percent, and an adjusted figure of 4.79 billion yuan (≈ US$675M / HK$5.27B). Part of that comes from cost discipline; part from the mix shifting toward transaction fees, which carry better economics than advertising listings.
The credit business is the part regulators watch. The company has been moving toward a lighter model with bank partners, capped loan pricing at 26 percent and reported a 90-day delinquency rate of 2.9 percent in the fourth quarter. Lending to owner-operators is a business where a good matching model is not obviously an advantage — a driver who cannot find a backhaul is a driver who misses a payment.
There is also an edge AI does not solve. The platform's value depends on both sides showing up: shippers post because trucks are there, drivers wait because loads are there. Autonomous trucking, which the company entered by taking a majority stake in Plus in 2025, is a hedge against a future where the driver side of that market thins out.
What readers can do now
- If you ship freight in China, ask platforms for fulfilment rate and on-time rate on your lane, not the national average; the difference between a 40 percent and a 70 percent lane is your expediting cost.
- If you drive for a platform, treat the "AI assistant" as a scheduler with the platform's interests baked in: check the recommended load's total revenue per kilometre, including the empty run to pickup, before accepting.
- Watch the fulfilment rate, not the order count. Order growth can come from reposting; fulfilment growth cannot.
Honest limitations
The matching-efficiency and empty-distance figures (40 percent and 27 percent) come from a company vice-president quoted by Guizhou provincial media and were not independently measured; we have no methodology for either. The 42.7 percent fulfilment rate, 63.9 million quarterly orders, 3.28 million monthly active shippers, 4.59 million active truckers, 88.6 percent commission penetration and 26.3 yuan (≈ US$3.70 / HK$28.9) per-order monetisation are company-reported from its 2025 results. The "three empties" decline and the claimed cumulative economic benefit of 380 billion yuan (≈ US$53.5B / HK$418B) are modelled figures published by the company and provincial media; the savings figure has no auditable basis. The production-agent description comes from a conference talk, not a product disclosure. Autonomous trucking is an early-stage investment, not a deployed business.