As generative AI rapidly transforms from a novelty into a necessary tool, it presents extraordinary opportunities for growth and improvement in logistics – but only if freight companies can feed their AI language models with relevant data, says Frank Kenney of Cleo.
In order for AI to be useful to the supply chain, it has to use data that’s specific to the goals the organization wants to achieve. The model used by an LTL firm based in Baton Rouge will differ from the model used by a freight brokerage in Madison, Wisconsin.
Kenney poses a thought-provoking question: what role will AI play in the world of logistics in ten years?
Generative AI won’t likely reach a plateau in logistics for quite some time, even with current technology. “In the meantime, we have to figure out integration to get the data AI needs,” Kenney said.
Without custom modeling and highly specific data, insight from generic language models aren’t helpful to everyone.
“It’s great to feed aggregated data into AI and get a prediction, but how useful that prediction is will depend on whether or not it’s tailored to my business,” Kenney said.
Even within the same sector of the freight industry, different companies might depend on entirely different models for their data, and that’s to say nothing of the diverse variety of roles each company plays in the supply chain.
Safety and compliance departments, brokerages, freight auditors, fleet managers and yard operators, for example, will all require vastly different information and contexts for their analytics and tracking systems.
“Right now, AI is hampered mostly by data gathering,” Kenney said. “Data gathering is an integration issue. How do I integrate with visibility platforms, brokerages, warehouses, 3PLs, and every other organization I work with? That’s a question we’ll all have to ask.”
According to Kenney, integration is the big choke point or bottleneck for getting usefulness out of APIs and AI in general. Insights and recommendations are only meaningful if they’re based on accurate and specific information.
Use cases for generative AI are inherently differentiated by company. What works for one sector doesn’t work for another.
Take, for instance, a household Alexa or any other voice-controlled digital assistant.
“We all have something like an Alexa now,” Kenney said. “Digital assistants control different things for different people. What my house calls ‘Dad’s light’ is the light in my office, but saying ‘Dad’s light’ to any other device in the world would not only be useless, but probably counter-productive as well,” he said.
The input, context, and level of integration matter in any system, whether it’s within a household or across an industry. “If another household means something else by ‘Dad’s light,’ you’re going to end up with a different result from what you expect,” Kenney said.
The only way to consistently get relevant data, according to Kenney, is integrating with your whole ecosystem.
“How much do I need to integrate in order to be useful? That’s the question we’ll be working on for the next 10 years,” Kenney said. “There will be diminishing returns as you increase the percentage of how much you automate and integrate.”
As AI continues to advance and embed itself in operations over the next decade, logistics companies need to figure out the best approach for maximum ROI.
“There will be expenses for every development, so it’s a matter of determining the most efficient integrations and testing them to determine the bare minimum of what we need,” Kenney said.
This process, says Kenney, will be a matter of prioritization and streamlining integrations with the necessary partners. “One important question every company in transportation will have to ask is, ‘who do I integrate with?’” Kenney said. “Some partners use only certain frameworks that may or may not be helpful.”
Even with integration between platforms and sectors, there still remains the question of how to gather all the necessary data consistently.
“Even if you get all the data you can possibly get your hands on, how do you clean it all up for the right format?” Kenney said. “We’ve had 30 years to get to 30 percent integration, and now it looks like we’ll have to get that number up above 85 percent in the next ten years to optimize AI,” he said.
“Most people working in the freight industry don’t have a thorough understanding of three or four dependencies down in their supply chain,” Kenney noted. “Nor do they know the end consumer, in a lot of cases. You might know the sphere you work with directly, but do you know where the raw materials come from?”
For instance, Kenney says, one 3PL may depend on a shipper who depends on multiple manufacturers who depend on vendors who depend on suppliers who depend on mines or chemical laboratories around the world, and all of those dependencies produce data points that could wind up being useful in analytics and predictions.
“As we slowly work toward making use of all of that data, we have to figure out how much to integrate and what makes the most sense,” Kenney said.
“This massive integration is a mountain we have to climb to get to that promise of what AI will ultimately mean for logistics,” Kenney said. “It will take a lot of incremental development, and we have the next decade to understand the supply chain intimately.”
According to Kenney, tech giants like Microsoft and IBM will win the AI arms race, so that’s not where logistics software developers or 3PLs need to focus their energies.
“Let’s start thinking about the quality of data and input before we get caught up in the hypothetical potentials of automation,” Kenney said.
Click here to learn more about Cleo.
The post Integration will cure the bottleneck for AI in logistics appeared first on FreightWaves.
