C.H. Robinson: AI agents for freight quotes, orders and tracking

C.H. Robinson receives tens of thousands of customer emails a day, and routine requests such as price quotes could wait hours before an employee picked them up. Each shipper has its own steps and preferences, so automating the work meant capturing customer-specific knowledge that previously lived with individual staff.¹

Microsoft reports that the company built its own generative AI tools on Azure AI Foundry (now Microsoft Foundry) with Azure OpenAI in Foundry Models, Azure SQL Database, Azure Cosmos DB and GitHub. The tools classify incoming email by freight mode, pull out the details, find missing information and then complete the steps a person would take, with a human feedback loop. Email price quotes went live first; shipment orders, pickup and delivery appointments and driver status updates followed, over about 12 months.

In April 2025 the company described a wider set of agents for quotes, order processing, sourcing truck capacity, appointments, checking loads in transit and tracking updates, plus a truck posting agent launched that January. That release does not name Microsoft, so the link between the later agents and Microsoft services rests on the 2024 Microsoft story.¹²

In our view, C.H. Robinson shows a sensible sequence for brokers and forwarders: start with the highest-volume, most standardized email task (quotes), keep a person in the loop, then extend the same pipeline step by step across the shipment lifecycle. All figures come from the company, the productivity numbers have no published method, and the Microsoft story is from 2024, so the scale-up evidence comes from the company's own release. Firms with lower email volumes may get most of the gain from structured web forms or a single extraction model before building agents.