ImplementationAdoption storyFor: Supply chain and procurement · Industry operations

C.H. Robinson AI agents: 1,306 fewer staff, more freight, pay bill up 3.1%

C.H. Robinson's Q3 filing credits AI agents for 10.4% fewer staff on 3.5% more truck freight. Pay costs still rose 3.1%: plan on output per person, not payroll.

At a glance

Company
C.H. Robinson · Logistics and transport · 5,000+ employees
Team and use case
Supply chain and procurement · Freight quoting, order entry, tracking and carrier payment
Tools
Custom-built AI agents (models and vendors not named)
Stage
Scaled across teams, since Not dated; the company measures from the end of 2022
Results
  • Average employee headcount, Q3: 12,559 (2025) → 11,253 (2026)
  • North American truck and LTL volume, Q3: Up 3.5% year over year
  • Personnel expenses, Q3: $349.3 million (2025) → $360.3 million (2026)
  • Productivity, company's undefined measure: Up more than 15% year over year
Cost
Severance and related personnel costs of $30.4 million in the first nine months of 2026. Cost of building and running the agents: Not disclosed.
People
Average headcount down 1,306 in a year; the split between layoffs and unfilled vacancies is not disclosed. Incentive pay rose for those who remain. Who checks the agents' output: Not disclosed.
Evidence
Company-measured, method disclosed

The company measured its own result and says how: the baseline, the period and what was counted.

What this leaves out: Headcount, volume and expense figures come from an unaudited earnings release filed with the SEC, with year-earlier baselines. The productivity percentages have no stated method, and the link from AI agents to lower headcount is the company's own attribution.

C.H. Robinson, the freight broker, ran the third quarter of 2026 with an average of 11,253 employees, 1,306 fewer than a year earlier (down 10.4%), while its North American trucking unit moved 3.5% more shipments, according to the earnings release it filed with the US Securities and Exchange Commission on 28 October 2026. The company credits "custom-built AI agents". Its personnel expenses rose anyway, by 3.1% to $360.3 million.

What the agents do

The investor presentation filed the same day lists eight kinds of agent: quoting customers, building and updating orders, contacting carriers for tracking updates, booking and rescheduling delivery appointments, posting available trucks, recommending loads to carriers, collecting carrier documents, and paying carriers. It calls them "a sample of the agents performing tasks that defied automation for decades", and gives reading, quoting and entering orders from email as examples. The filing does not name the models or vendors behind them.

The numbers, by unit

  • North American Surface Transportation, the truckload and less-than-truckload brokerage: average headcount fell from 5,187 to 4,597 (down 11.4%) while volume rose 3.5%. That is about 17% more shipments per employee (1.035 ÷ 0.886 = 1.168).
  • Global Forwarding, the ocean and air business: average headcount fell from 4,245 to 3,573 (down 15.8%) and operating income rose 63.3% to $80.0 million. The presentation also credits higher ocean and air freight rates.
  • The company's own claim: productivity in both units improved by "greater than 15%" year over year and "more than 65% since the end of 2022". Neither document says how productivity is measured.

The catch: the pay bill went up

Personnel expenses were $349.3 million a year earlier and $360.3 million this quarter. The release says the rise was "primarily due to higher incentive compensation reflecting our strong operating performance", partly offset by cost cuts and productivity. Per average employee that is about $27,800 a quarter then and $32,000 now, up 15% ($349.3 million ÷ 12,559; $360.3 million ÷ 11,253).

Leaving cost more too. Severance and other personnel expenses "related to workforce reductions" were $30.4 million in the first nine months of 2026, against $14.8 million in the same period of 2025.

Operating income rose 23.1% to $271.8 million. The release attributes the growth in gross profit primarily to higher gross profit per transaction (up 17.0% per truckload shipment) and higher volume. Wren's read: the filing shows more freight and more margin per person, not a smaller payroll.

What the filing does not say

It gives no cost for building or running the agents, no error rate, and no account of who checks an agent's quote or carrier payment. It does not say which roles went, or how many of the 1,306 left through layoffs rather than unfilled vacancies.

The model is about to be applied to more people. C.H. Robinson announced on 5 October a deal to buy the broker RXO, expects to close in the first half of 2027, and estimates about $300 million a year of net cost savings, 80% of it (about $240 million) from "the Lean operating model and fleet of AI agents deployed across workflows".

What to do if you run freight or order operations

  • If you run logistics, order or carrier operations: count shipments or orders per person per month now, for quoting, order entry, tracking and appointment booking separately, so any automation is judged against a baseline you set and not a vendor's.
  • If you are the finance partner on the business case: carry headcount, incentive pay and severance as three lines. C.H. Robinson cut 10.4% of average headcount and its personnel expenses rose 3.1%.
  • If you buy brokerage from C.H. Robinson or RXO: ask your account manager which of your quotes, appointments and tracking updates an agent now handles, and who you call when one is wrong.
Was this useful for your work?

Take this to work

Forward this

C.H. Robinson's 28 October 2026 SEC filing shows average headcount down 10.4% in a year while North American truck volume rose 3.5%, which it credits to AI agents it built for quoting, order entry, tracking and carrier payment; its personnel expenses still rose 3.1%. A business case for automating our freight or order desk should be built on shipments per person, with incentive pay and severance costed separately.

Questions to ask

  1. How many shipments or orders does each person on our desk handle a month today, by task, and who owns that number?
  2. What would an agent for quoting or order entry cost to build or license and to run at our volume?
  3. Who reviews an agent's quote, appointment or carrier payment before it goes out, and how are errors counted?
  4. If volume per person rises, does our incentive plan pay out more, and is that in the forecast?
  5. Which roles change or go, and by attrition or by layoff, with what severance?

The arithmetic

Change in shipments per person = (1 + volume growth) ÷ (1 + headcount change) − 1. Change in pay bill = personnel expenses now ÷ personnel expenses a year ago − 1.

C.H. Robinson's North American unit: volume up 3.5%, average headcount down 11.4%, so 1.035 ÷ 0.886 − 1 = about 17% more shipments per person. Company-wide pay bill: $360.3M ÷ $349.3M − 1 = up 3.1%. The cost of the agents is not stated.

Questions this article answers

How much has C.H. Robinson cut headcount with AI?

Its third-quarter 2026 earnings release reports average employee headcount of 11,253, down 10.4% from 12,559 a year earlier. The company ties its productivity gains to AI agents it built itself, but does not say how many departures were layoffs.

What do C.H. Robinson's AI agents do?

The company's investor presentation of 28 October 2026 lists agents that quote customers, build and update orders, chase carriers for tracking updates, book appointments, post trucks, recommend loads, collect carrier documents and pay carriers.

Did C.H. Robinson's payroll costs fall after automation?

No. Personnel expenses rose 3.1% to $360.3 million in the third quarter of 2026, which the company attributes mainly to higher incentive pay, and it recorded $30.4 million of severance and related costs in the first nine months of the year.

How does C.H. Robinson measure AI productivity?

It reports improvements of more than 15% year over year and more than 65% since the end of 2022 in its two main units, but neither the earnings release nor the presentation defines the measure.

Send this to someone who runs a teamLinkedInXBlueskyHNEmail