Uber Freight used to have people review almost every document behind a carrier payment. Now AI checks each proof of delivery against the load record, and when every check passes, the payment goes out without anyone touching it. Its people now focus only on the complex or escalated cases. Uber Freight says this cut disputes in half and moved 2,500 hours of its experts' time from checking paper to solving problems. You don't need Uber's budget to copy the core idea. Below is what they did, what stayed with people, and a smaller version a broker or 3PL with 50 to 200 people can start with.
What Uber Freight changed
In March 2026 Uber Freight announced DocAI, a system that checks carrier paperwork before payment. Before it, the team was reviewing hundreds of thousands of documents every month. The company described the old workflow to Food Logistics as one that "required human review for nearly every document involved in the payment process," from the proof of delivery (POD, the signed paper that shows the freight arrived) to lumper fees (paid to the crew that unloads a truck) and detention (pay for a driver kept waiting).
The goal was to flip the default. Instead of every load starting as "manual review required," a load with clean paperwork starts as "auto-approve." Uber Freight calls the result a touchless payment. A document is uploaded, the system verifies it, and payment is triggered with no person in the loop.
That sounds like a small wording change. It isn't. In most back offices the default is still that a person opens every POD before anyone gets paid, so the speed of carrier pay depends on how fast that person gets through the pile.
How the checks work
The steps are easy to describe. The hard part is doing them reliably on documents that arrive in every shape and quality.
- The system works out what kind of document it is, such as a POD or a bill of lading.
- It pulls out the key details, like the PO number, the delivery date and the address.
- It compares those details with the shipment record in Uber Freight's own system.
- It checks the softer requirements, like a stamp or a signature a specific customer asks for.
- If everything passes, payment is triggered. If something is wrong, the carrier is told right away, and the hard cases go to Uber Freight's experts.
One detail matters more than it looks. Uber Freight doesn't rely on a single AI model for this. It runs several large language models together (the kind of AI behind tools like ChatGPT), which the company calls a "council of LLMs." Its CTO, Val Marchevsky, put it plainly in a FreightWaves interview in July. "There isn't one model that does everything."
The numbers Uber Freight published
These are the company's own figures from its announcement. I haven't seen an outside audit of them, so read them as what Uber Freight reports, not as a result any broker will get.
- More than 99% accuracy when pulling key data out of documents.
- Downstream disputes cut in half, because billing errors get caught before the invoice is generated.
- Every document audited in full.
- 2,500 hours of expert time moved from checking documents to solving problems.
- Carriers told about a problem right away, where Uber Freight puts the industry norm at 12 to 24 hours.
In the FreightWaves interview, Marchevsky said document work like PODs and bills of lading gave Uber Freight its clearest early proof point for AI, and that it didn't come easy. He said it has eliminated more than 50% of the manual work the process once required.
The number I'd watch isn't the accuracy. It's the disputes. A billing dispute means calls, emails, a late payment and sometimes a customer or carrier who stops working with you. Catching the problem before the invoice even exists is where most of the money is.
What stayed with people
Uber Freight didn't take people out of payments. It moved them. The company says its experts now spend their time "only on complex or escalated issues" instead of opening every POD.
It also didn't automate everything at once. When it announced the system, the focus was carrier PODs, and the company said it plans to expand to other document types, including lumper receipts and invoices. That's a large company choosing to start narrow and get one thing right first.
I think that's the right split. A machine is good at checking that the PO number on the paper matches the PO number in the system, over and over, without getting tired. A person is better at the conversation with a carrier when something went wrong at the dock.
The smaller version a broker can copy
You don't need a council of models or a room full of engineers to get most of the benefit. You need clear rules and a safe way to switch them on. This is how I'd start at a broker or 3PL with 50 to 200 people.
- Write down what a clean POD means for you, such as the load number, delivery date, consignee name, a signature, any stamp a customer requires and no notes about shortage or damage.
- Start with PODs only and leave lumper receipts and other extra charges for later, the same way Uber Freight started with PODs.
- Run the checks on last month's loads first, without paying anyone, and compare what the system would have approved with what your team actually did.
- Turn on auto-approve only for documents that pass every check, and send everything else to a person with the reason it failed written at the top.
- Tell the carrier the moment something is missing, so they can fix it while the details are fresh.
- Track three numbers every week, the share of loads paid with no touch, the number of disputes and the days from delivery to payment.
Most brokers already have the data for this in their TMS (transportation management system, the software that holds each load). The load record is what the paperwork gets checked against, so it has to be right. If the delivery date in your system is wrong, a perfect POD will fail.
Where this goes wrong
- Wrong load data makes good paperwork fail, because every check compares the paper with your system.
- Customer rules that live in one person's head never make it into the checks, so write them down before you build anything.
- Switching on automatic payment before you've measured is the fastest way to lose trust, so test quietly on old loads until the misses are rare and you understand each one.
- A document can look right and still be wrong, so keep a person on new carriers and unusual loads until you have some history.
- If nobody owns the failed documents every day, carriers end up waiting longer than they did before.
The useful lesson from Uber Freight is the default. Most brokers still start every carrier payment with a person checking the paperwork. Pick one document type, write the rules down, test on past loads, and let people spend their time only on what fails. If your back office spends its week opening PODs and chasing signatures, that's the kind of system we build at OrbiResearch, with a person in charge of every exception.
Work with us if you want a small first version built and tested on your own loads.