Many of the products you’ll use today were made by people all over the world, and getting them from the final factory into your hands was likely a complex process involving more than a dozen companies, a variety of government agencies, and lots and lots of paperwork.
In 2013, Ryan Petersen founded freight forwarding company Flexport to bring modern tech solutions to the cumbersome — and often-archaic — trade industry, and today, his company is helping thousands of brands save time and money as they move their goods around the world.
“When you’re moving cargo from door to door, you’ll have 15 or 20 different companies that will do something on that container … We’ve built software to integrate with all these parties, reduce the amount of manual work that’s required, and improve the quality quite a bit,” he told Freethink.
“Generative AI is huge for us.”
Alex Nederlof
Artificial intelligence has played a key role in Flexport’s success — the company uses it for price estimates, route optimization, risk analysis, and more, Alex Nederlof, Flexport’s director of engineering, told Freethink.
Most recently, Flexport has been looking for — and finding — ways to incorporate generative AI (the branch of AI focused on training systems to be able to generate text, images, videos, and other types of content) into its operations.
“Generative AI is huge for us,” Nederlof told Freethink. “We have fully embraced it in a number of areas.”
Freethink recently got a chance to talk to Flexport about how it’s using generative AI, the areas it doesn’t expect will benefit, and the “holy grail” of applications that it is in the process of building.
Document digitization
Shipping involves a ton of documentation, and parsing it all is one of the biggest challenges in freight forwarding. It’s also a challenge that generative AI is helping Flexport overcome.
“Documentation is mind-numbing, uninteresting work that requires no skill and is easy to automate with this new technology,” said Nederlof. “It’s work that nobody wants to do, but it’s critical to do it, and it needs to also be done in time.”
Today, much of the documentation that Flexport receives is in PDFs or spreadsheets attached to emails, but some organizations still fax or mail documents to the company that someone then has to scan into a computer.
That’s just the first step. Once the documents are in a computer, Flexport then needs to actually parse the information, and because much of the data it receives is unstructured — not fitting into any sort of pre-defined, easily searchable format — traditional AI hasn’t been terribly helpful in this department. That meant the company would need to have people available 24/7 to read documents, make sense of them, extract the relevant information, and then take any actions necessary as quickly as possible.
Flexport is now using multimodal large language models (LLMs) — generative AIs like ChatGPT that can understand images or language and generate text — to help it quickly understand what kind of information is included in a document and what needs to be done in response to it.
“We use [LLMs] to automatically digitize commercial invoices, bills of lading, emails the customer sends us: ‘Hey, I’m a day late with my truck cargo.’ We can automatically parse it and update our track and trace [system] and our status internally,” said Nederlof.
Because LLMs make mistakes, Flexport still needs to have a person review the AI’s recommendations, but having it “do the leg work” makes the process much faster, according to Nederlof: “It saves a tremendous amount of time, but more importantly, it’s 24/7.”
AI calling
Trucking is an essential part of the supply chain, accounting for an estimated 50% of the money spent on global trade. It’s also another area where Flexport has found a way to use generative AI to improve operations.
“Just about every commodity that gets moved from one location to another touches a truck at some point in the shipment lifecycle,” explains Flexport. “Trucking is at the start and end of every air and ocean shipment, as well as being a standalone transportation mode all by itself.”
“The bots are always friendly. They’re not in a rush. They’re not on a clock.”
Alex Nederlof
Communicating with truckers is vital to getting goods where they need to go as quickly as possible, and generally, the safest way to do this is through phone calls — you don’t want truckers trying to read text messages or navigate an app while on the road.
“Sometimes truckers don’t have their papers filed, the cargo is moved, or something else, and we need to call them,” Nederlof told Freethink. “We used to have a call center for this. We now use AI to do that, and truckers actually seem to like it better.”
That positive reaction was the most surprising part of the deployment, for Nederlof.
“Robocalling has a pretty bad rep, and I didn’t really see how you could like it more than talking to a human, but when you hear these conversations, the bots are always friendly, they’re not in a rush, they’re not on a clock to do X amount of calls per hour,” he told Freethink.
Flexport can even personalize the bot’s dialect, which can make talking to the AI feel like talking to a neighbor.
“They can have an interactive conversation,” said Nederlof. “They can talk about anything, basically — the bot is very capable — and we notice that the truckers take their time. It’s a very pleasant experience for them, and for us, we no longer need a huge call center to do this.”
Flexport is now working to roll out the AI calling to other facets of its operation to improve communication with manufacturers, vendors, and others along the supply chain. It’s also looking into bots that could serve as translators between people.
“If you’re an American trying to call an Asian factory or if you’re a Dutch person like me trying to call somebody who’s French, there’s a lot of barriers that we currently have that will dissipate because of this technology,” said Nederlof.
Expert agents
There’s a reason 99% of companies that buy, sell, or ship goods internationally use freight forwarding companies like Flexport rather than trying to manage the movement of their products along the supply chain themselves: logistics is complicated.
“It’s dense and wide,” Nederlof told Freethink. “It’s not physics, but every country has their own regulations that don’t necessarily … make sense because they were made by humans and generations of political entities.”
Because of this, making even a seemingly straightforward change — like moving manufacturing from the Philippines to Vietnam to take advantage of better duty rates — can have ripple effects that a company needs to fully understand before it can know whether the move is beneficial.
“The way you do your ocean freight might differ. The way you have to do your customs is different. You have to get certain trade agreements,” explained Nederlof. “These people now need to get 15 different experts into the room to answer the question, ‘Is this hypothetically possible?’ not even going into the details.”
“That’s like the holy grail for us.”
Alex Nederlof
In the long term, Flexport hopes to use generative AI to build “knowledge agents” capable of sharing the information customers would typically get from those experts.
“The AI has access to all your data, and it can translate what you’re asking into things that the database understands — it can talk to the database, basically,” said Nederlof.
“Then we have all this knowledge around the world — we know every trade agreement, every ruling, every regulation, every ocean route, every air route, every airport,” he continued. “These agents can combine these things.”
Flexport is experimenting with knowledge agents focused on certain aspects of shipping — there’s one that knows all about customs, for example, another that’s an expert in the latest freight news — but the ultimate goal is to combine them into a single agent that’s an expert on everything involved in freight forwarding.
“That’s like the holy grail for us: to get there as fast as possible and unlock that for our clients,” said Nederlof.
The big picture
Generative AI may be useful for a range of applications, but it does have its limitations.
“I’ve not seen areas where I’m 100% sure AI won’t be able to do it, but there are areas where I see it’s a lot more work to get the AI to do the right thing … Basically anything with math and numbers, you need to do a lot of work to get that right,” said Nederlof.
Between all the things that generative AI can do and all the ways that Flexport uses more traditional AIs, Petersen believes the company will be able to automate about 50% of the manual work that takes place in a freight forwarder by the end of 2025.
“That is ‘manual’ not in the sense of lifting freight, but of transposing data, making phone calls, and workflow to coordinate freight movements,” he told Freethink. “That’s a really big deal. It probably translates to about a 5% decrease in the cost of shipping things around the world.”
EDITOR’S NOTE, 11/27/24 at 4:40 pm ET and 12/5/24 at 4:00 pm ET: This article was updated to clarify how the dialect feature works for AI calling and the status of Flexport’s knowledge agents.
We’d love to hear from you! If you have a comment about this article or if you have a tip for a future Freethink story, please email us at [email protected].