Practical AI tools in transportation management systems

Route optimization

An ideal route transports a load to its destination quickly, safely, and efficiently.

Traditional route planning methods often use static maps to find the best path—but speed, risk, and efficiency are affected by much more than map data.

In mathematics, an optimization problem is based on this exact situation: the vehicle routing problem, introduced to the field in 1959 in a scholastic paper called “The Truck Dispatching Problem”—go figure.

Researchers have spent decades teaching computers to find the best feasible routes. Like with many other optimization problems, machine learning techniques are now improving real-world vehicle routing. AI can greatly improve less-than-truckload routing in particular, with its significantly more stops and possible solutions, as researchers at MIT showed in 2021.

“The first and classic use case of AI in LTL has been around optimizing the routing of shipments to trucks,” CLI’s Wiesen said. “The reason why it is more than a math model—the reason it is AI—is because the input to the model involves travel time.”

Machine learning applications are particularly suited for juggling location, time of day, local traffic patterns, weather, seasonality, and more to estimate travel time. To AI’s advantage, much of that additional data is plentiful, making route optimization a perfect fit for the technology.

Many TMS route optimization tools connect directly to prominent technology companies, such as Microsoft, for the heavy lifting of data gathering and computation.

Text recognition

While industry groups are moving toward standardized digital forms such as electronic bills of lading, paper forms are still common.

“Not everyone is doing that yet, and so there’s still a tremendous amount of paper,” CLI’s Wiesen said. “Historically, the next step was someone’s fingers would be on a keyboard, and they would be transcribing the data from that piece of paper into a computer system. It was low-value, high-touch work.”

Manual data entry is slow, expensive, and prone to error. Many carriers use solutions to expedite the process, but human review can still be needed.

Companies digitizing their paperwork can use scanners and software to convert the written information into digital data. The dominant solution is optical character recognition (OCR), a powerful and ubiquitous approach that is present in almost all text processing applications. Companies have used OCR for over 50 years, and the technology has become most accessible with the advent of cloud computing.

OCR programs can recognize characters and words with great accuracy, but often can’t contextualize those characters without laborious, explicit guidance. Modern OCR has utilized machine learning techniques for over a decade, but new AI developments are further optimizing document processing: Large language models (LLMs), the power behind tools like ChatGPT, can ease the context problem.

“What AI has done is it’s given us the ability to contextualize that data,” Wiesen said. According to him, a perfect example of LLMs in action is identifying zip codes. “If you see a five-digit numerical value after a two-letter abbreviation that’s clearly a state, then you know it’s a zip code. We just know that inherently, because we’re human and we’re sort of smart. The computers now contextualize data the same way.”

How CLI uses AI

Carrier Logistics primarily serves asset-based LTL motor carriers with its TMS platform, FACTS. According to Wiesen, CLI sees itself as a cutting-edge platform with progressive customers. Fitting that description, the company offers several AI tools. The FACTS platform uses machine learning to optimize LTL operations, acquire shipping location information, weigh debt risks, and parse documents.

Automated shipping location information

CLI uses AI to automatically provide detailed information about new shipping and receiving locations. It calls this solution LOC-AI, or Location Management Artificial Intelligence.

CLI rolled out LOC-AI in 2020, initially to help address changing shipping needs during the pandemic. In LTL, trucks frequently visit new, unknown sites to distribute parts to one-off locations.

“If I don’t know the attributes of a stop, it’s very hard to do optimization. Maybe there’s a particular location that requires a liftgate that I shouldn’t go to in certain hours because they’re in a restaurant zone,” Wiesen said. “The LOC-AI product gives them information about a location they never serviced without them having to spend time going to Google Maps and looking each one up, trying to decide what they think.”

The LOC-AI tool taps into large data sources and uses heuristic models to determine if the location requires special equipment, extra time for processing, and more.

“[They should do] anything they can to ensure they have the right equipment [and] the right driver,” Wiesen said.

AI debt risk scoring

The company’s accounts receivable risk analyzer provides automated debt risk scores for accounts. The A/R risk analyzer uses AI to identify at-risk accounts. Rolled out in 2023, Wiesen describes the feature as “analysis of money that customers owe in order to try and calculate risk of each account—risk of default, of not getting paid, of having bad debt.” The system evaluates risk through debtor behavior. A consistent debtor, always paying on time, is not very risky, he explained.

“It’s the one that maybe never owed me a lot of money and suddenly does, or the one who had been paying me reliably but suddenly is paying me a little bit slower,” Wiesen said. “It’s the accounts that have variability and change in behavior where we see a lot of inherent risk.”

Automated data entry

CLI recently added AI-assisted data entry to its platform. The solution pairs large language models with text recognition programs to reduce the need for manual data entry. “With AI, we’re now able to extract data from the documents using vision-type AI products,” Wiesen said. “It’s able to contextualize the data that it’s pulling off of documents to understand what each of those data elements is, and then it can normalize and inject it into our TMS.”

What should fleets look for?

With revolutionary technologies, failed initiatives, and deceptive marketing waiting around every corner. Carriers that want to learn the best uses for AI should be careful to minimize error risk, prioritize safety, and keep an eye on projected costs and benefits.

Safety