AI agents move into the freight back office

AI agents can now audit your carrier invoices. Ask what they recover, not what they approve.
On 14 October, New York startup Evos announced a $9 million pre-seed round for AI software that does freight back-office work. Pre-seed means very early-stage money. Uncork Capital and Field Ventures co-led the round. Crosslink Capital, Utopia Capital Management, Sandwith Ventures and angel investors also took part. Evos calls its products “autonomous operators.” They audit carrier invoices, build loads, follow up on paperwork and check data against a firm’s transportation management system, or TMS, the software forwarders and brokers use to run shipments.
A day earlier, fleet-software group Samsara made its AI agent builder available to all its customers. It reported that use of its agents is growing quickly, and that voice calls, mostly placed by the agents themselves, now outnumber chat.
Our view is that AI in freight has moved past the chatbot stage. It is now doing clerical work that used to belong to a desk of coordinators and auditors. The proof offered so far, though, measures how often humans agree with the machine. It does not measure how much money the machine saves. Forwarders should keep those two apart.

What Evos is selling

Evos targets small and mid-sized industrial businesses. Forbes reports these typically have 20 to 200 employees and work in logistics, manufacturing and energy. Founder and CEO Urav Shah told Forbes that “This segment of the market has been neglected.” He said these firms are “stuck with legacy systems that require staff to spend much of their time on tasks such as data entry.”
The pitch is that the software learns by watching experienced staff. In Shah’s words to Forbes: “The most capable system in many of these businesses is the person who has been there 20 years.” He said an Evos operator “learns the job from these people, improves on it, and proves it can match them” before it takes on any work by itself.
Supply Chain 24/7 describes a cautious rollout. At first, every decision goes to the customer’s team for approval. Tasks run unattended only after the software’s decisions consistently match the team’s, and unfamiliar cases go to a person. Each customer gets a model trained only on its own data. Shah told Forbes that “An Evos operator is live within 24 hours.”
Evos says it passed $1 million in annualized recurring revenue (ARR) four weeks after going live. ARR is a run rate: current monthly contract revenue multiplied by twelve. It is not cash collected. Forbes reports that Evos has about 20 customers and “believes its rapid growth will see it hit $10 million of ARR within six months.”

The numbers behind the invoice audit

The freight-specific evidence comes from one customer, a freight brokerage. According to Supply Chain 24/7, an Evos operator audited LTL carrier invoices against the brokerage’s TMS. LTL, or less-than-truckload, is freight that fills only part of a trailer and shares space with other shippers’ goods. LTL bills are known for disputes over weight, freight class and extra charges called accessorials.
“Evos said the operator made nearly 3,500 audit decisions over two months,” the outlet reports. “According to the company, human managers accepted 97.3% of those decisions exactly as proposed.” By our arithmetic, that leaves roughly 95 decisions that a human changed. The operator “also closed out more than $600,000 in invoice value over five weeks, according to Evos.”
Then comes the key caveat. Supply Chain 24/7 states that Evos did not report how much the audit saved or recovered.
That gap is our main point. A 97.3% acceptance rate shows the software agrees with the brokerage’s managers. That is useful, because it means fewer hours spent on routine checks. It does not show that the software catches overcharges the managers would have missed. If the old process let bad invoices through, a tool that copies it closely will let them through too. And “$600,000 in invoice value closed out” counts how much money was processed. It is not the amount clawed back from carriers.

Samsara shows where agents are actually being used

The Samsara news shows the same shift from a different direction. FreightWaves reported on 13 October that Samsara made Agent Studio, its no-code agent builder launched in June, available to all customers, included in existing subscriptions. Its latest 500,000 agent interactions took 16 days, FreightWaves reported, down from 112 days for the first 500,000. That is about seven times faster. More than 1,000 organisations use Samsara agents weekly.
What stands out is how the agents are used. Voice passed chat in interaction volume within four weeks of launch and has led every week since. Most voice interactions were calls that the agents placed themselves. Samsara says they can call drivers with instructions when trucks arrive at unfamiliar facilities. Samsara chief product officer Johan Land said “AI is quickly adopted when it’s useful, and in physical operations that means an agent has to understand the operation” and meet people where the work happens.
FreightWaves also noted what Samsara left out. The company did not say how many customers use agents in production or how many have built custom agents. It also did not provide detailed measurement methods for all the customer results it reported, which include an 84% cut in weekly manual review time at S. Bertram Foods.

The wider picture: money is flowing, results are not yet

Investors are clearly backing the theme. Forbes lists Augment, HappyRobot and Envoy AI as logistics-focused rivals to Evos. In August, Dealroom reported that HappyRobot raised a $150 million Series C at a $1.2 billion valuation. HappyRobot began in logistics, and its agents negotiate with truck drivers and schedule appointments. It counts DHL and Kuehne + Nagel among more than 150 enterprise customers. So two of the world’s largest forwarders already buy agent software of this kind.
Results across the industry are much thinner. FreightWaves cites a March 2026 BCG survey that “found that about 40% of logistics service providers had deployed AI beyond pilot programs,” “but only about 10% had scaled it across core operations.” It adds: “Just 13% reported measurable value from AI.” A September 2026 Descartes survey, also cited by FreightWaves, found that 19% of shippers and 15% of logistics service providers were using AI at scale.
Freight tech has been here before. Digital freight broker Convoy raised more than $1 billion, according to PitchBook figures cited by Supply Chain Digest, and was valued at $3.8 billion in 2022. It told staff in an October 19, 2023 memo that it was shutting down. CEO Dan Lewis blamed a “massive freight recession” and said “market forces were too strong for us to withstand on our own.” The lesson for buyers is not that new freight software fails. It is that fast growth and big funding rounds do not guarantee the supplier will still be around to support the product.

What this means for forwarders

Carrier invoice audit is one of the clearest margin levers a forwarder or broker has. Every overbilled accessorial or wrong freight class that gets paid comes straight out of margin. Small and mid-sized forwarders, the 20-to-200-employee firms Evos is chasing, usually cannot afford a dedicated audit team. They pay invoices that a larger competitor would dispute. Done well, agent software could close that gap.
Here is how we would approach it:

• Ask for recovered dollars, not acceptance rates. Ask what share of invoices the agent flagged, how many disputes were won, and how much money came back from carriers.
• Run it alongside your current process first. Evos’s own approval-first model allows that. Compare what the agent flags with what your team flags over the same weeks.
• Check where your data goes. Evos says each customer’s model trains only on its own data. Get that in writing, along with what happens to the data if the vendor closes.
• Watch the cases the agent hands back. The unfamiliar cases that go to a person are often where most of the money is.

On the trucking side, Samsara’s voice agents matter most to forwarders that run their own drayage (short trucking moves to and from ports) or that use carriers on Samsara. Automatic calls to drivers about arrival instructions could cut missed appointments and detention, the fees charged when trucks are held too long, at warehouses and container freight stations. That depends on carriers switching the agents on. Forwarders don’t control that.
For shippers, the knock-on effect is gradual. If mid-sized forwarders audit more of their carriers’ bills, carriers will face more disputes on loosely documented accessorials. Over time, that should push more of those charges into clearer, written rate terms.

Our prediction

We expect the market to stop rewarding approval rates within about six months. By mid-April 2027, when Evos’s own six-month window closes, we think the back-office AI vendors winning new contracts with small and mid-sized forwarders will be the ones that publish audited figures for money recovered from carriers. We also expect Evos to fall short of its $10 million ARR target in that window, though we think it will keep growing. Invoice audit will be the first back-office task where many of these forwarders let an agent run without approving each item. That is because the results can be checked line by line against the money that comes back.

Sources

• \
• Supply Chain 24/7 — Evos Raises $9 Million to Automate Freight and Factory Work, 14 Oct 2026 \
• Forbes — Startup Evos Raises Funding To Target Industrial Businesses, 14 Oct 2026 \
• FreightWaves — Samsara makes AI agent platform generally available, reports 7x adoption growth, 13 Oct 2026 \
• Dealroom — HappyRobot raises $150M Series C at $1.2B valuation to scale enterprise AI, 4 Aug 2026 \
• Supply Chain Digest — Digital Freight Broker Convoy ran Out of Options and Money, Shutters Doors, 23 Oct 2023