C.H. Robinson executive highlights promise of generative AI in logistics
ATLANTA — With AI in general and generative AI in particular the talk of the FreightWaves Future of Supply Chain event, Megan Orth of C.H. Robinson reached for several analogies to describe generative AI’s use at her freight brokerage and in logistics generally.
“It’s kind of like APIs 10 years ago,” Orth said, referring to application programming interfaces, which allow different types of software to communicate with each other.
When APIs first emerged, Orth — senior director of commercial connectivity at C.H. Robinson — said she wondered, “Is this a buzzword? Is this really going to work?”
Reflecting on the decade since then, she said, “It did change our industry.” She cited real-time truckload rates and tracking appointments as productivity gains enabled by APIs.
“So when I think of AI, I think the same thing,” Orth told Daniel Pickett, FreightWaves chief data and technology officer, in a fireside chat Wednesday. “There are going to be very specific use cases that are going to help us crack some problems that we’ve just never been able to crack.”
Interns and swivel chairs
Another analogy Orth used for GenAI: “intern.” That’s the internal term at C.H. Robinson (NASDAQ: CHRW) that is often informally used to describe its capabilities.
“Think of this as your summer intern,” she said. “So you can go on vacation, but you’ve got to train it.” An intern can’t step in and do everything on day one, and neither can a generative AI tool, according to Orth; it needs to be trained.
“So that’s how we’ve been simplifying the message to our employees that this is something that will help you long term and that the intern will get really smart like an employee,” Orth said.
Orth also likened the technology to the swivel chair. She portrayed a typical C.H. Robinson employee as sitting in a swivel chair, moving to the left to get a piece of data, moving to the right to get a different piece of data and so on. “But you can use AI to bring this all together, and we’re trying to eliminate swivel chairs,” she said.
C.H. Robinson has been under pressure in its bottom line and its operations; it ousted its CEO right at the start of 2023, reportedly in part because company directors believe the 3PL’s technology was lagging.
New CEO Dave Bozeman has talked often about technology having a greater role at C.H. Robinson. That seemed to start paying off in the first quarter of this year when C.H. Robinson outperformed expectations and its stock price climbed in response.
Orth is at the front line of that effort, and her talk Wednesday was not just about generative AI’s theoretical uses but how C.H. Robinson has adopted it already.
Generative AI use at C.H. Robinson is “out of the lab,” she said, after “a lot of testing.”
She described a process in which testing certain applications of generative AI would begin with one or two customers and gradually ratchet up to 20. The customers whose use cases were part of the initiative were told that “you’re going to stay in this for three to four weeks, and then you give us a thumbs up.”
Orth said she considers C.H. Robinson “a front-runner in this space.”
“There are people who are getting gen AI to scale right, like us, and there are people who truly just don’t believe in it,” she said.
But Orth sought to drive home the point that generative AI is not just a technology. “A lot of it is behavioral,” she said, and that requires that it not be a technology initiative in which the company’s tech team is calling most of the shots.
“Make sure your business is represented,” Orth said. “If you just have your data scientists and tech team go in and do this, it’s not going to be successful. You have to have the different aspects of the business represented there.”
That helps ensure a proper “feedback loop,” Orth said. And that loop is something the developers at C.H. Robinson didn’t fully think about at the beginning of the process. “But it ultimately becomes important as we scale, because as we scaled, we needed that feedback,” Orth said.
That feedback can then be added to the generative AI, too. That process is not like changing code, Orth said. It’s more real time. “You need to keep feeding the model; we’ll deploy new models and then it will work,” she said.
More articles by John Kingston
Reducing waste, manual touches ‘big opportunity’ for C.H. Robinson
C.H. Robinson’s Q4 sees little improvement; shift at top of brokerage unit
Teamsters tie up with Staten Island Amazon union, win other elections
