Rusty Weiss used to sit and parse video of drivers for patterns in behavior to determine if any of those particular behaviors were predictors of near or real collisions.
Before applying AI to look for those connections, he said he would spend hours producing graphs and spreadsheets of data. But once machine learning was applied to that repetitive task, it allowed him to shift his time from analysis to prevention.
“We could have done it manually, but boy it sure did streamline where we needed to focus our attention to prevent those major collisions in the future, and it really helped test and validate training investment,” said Weiss, director of transportation services at Yahara Software, which provides data analytics and custom app development, among other things, to the transportation industry.
While AI isn’t new, in the past couple years alone, it has become a much more powerful tool. Technology vendors in the trucking sector have been investing in AI creation, and carriers are starting to climb on board in an effort to optimize operations at a time when budgets are stretched thin.
Seventy-five percent of fleets said they were considering investing in data infrastructure or AI tools in the next one to two years but had no specific plans yet, while 25% said they are still gathering information to make a decision in a Yahara poll taken during a recent webinar.
[RELATED: Five ways AI is transforming the trucking industry]
Many don’t know where to start, said Josh Peot, Yahara director of engineering. The first step, he said, is to collect your data from multiple sources and store it in one centralized location like a data lake. Then, build something that watches and processes that data.
“You can start small. You can start now,” Peot said. “… But the nice thing is, because you've stored that in its original form, if you want to go back and ask it a question later, you can. You own that data. It's in your sphere. You don't have to go request it from a vendor, or maybe they deleted it because they only store it for 30 days or something.
Then AI can be applied.
How and when to use AI
Peot said AI like machine learning can be used to automate anything that is repeatable work.
“If you're doing the same thing three times, you're probably going to do it a fourth time, maybe a fifth time. Is this something you could train a tool for,” he said. “There's this fear of it taking over people's jobs. It's more about augmenting people's jobs, so letting people do the people work and using AI to get rid of all the cruft you might have to do in a day, opening up your time.
“Instead of one person doing one … person's amount of effort, they're able to do five or 10 times that effort without any extra expenditure on their their side,” Peot added.
A layer deeper, generative AI can dissect data to find patterns and offer predictions, even allowing the machine to make decisions and create new content based trained data.
[RELATED: Geotab plans rollout of generative AI copilot]
It can also be used it to query your data like a large language model like ChatGPT, which is what most people think of when they think AI, Peot said. Say you have a 5,000-person company, and you’re looking for someone with a specific skill set. You can funnel your employees’ job descriptions into the data lake and query that data via AI to determine who that is.
It could also trigger alerts based on data. Peot said if you pull data in automatically, like weather information for example, AI could trigger alerts to drivers in an affected area or alert fleet decision-makers to determine if they need to delay a shipment.
Weiss said you can even use it to determine if one lane is more profitable than another based on data like profit margin and tractor specs.
“There are too many options,” Peot said.
Where to get AI
Weiss said there are many good off-the-shelf solutions carriers can start with, but “nothing beats actually working on projects.”
He said many off-the-shelf solutions need to be augmented based on your company’s needs because they may only meet 75% to 80% of your needs.
[RELATED: AI is launching the transportation industry forward]
Peot said start by giving the model general awareness of your data. If the model isn’t doing well, tweak it (train it) to behave better based on your needs.
But he offered one caveat to off-the-shelf solutions.
“Make sure you understand the licensing agreement,” Peot said. “Depending on your license tier and what tool you're using, there are different agreements on who owns that data and what that data can be used for.”
