Safety performance can make or break a trucking company’s reputation, not to mention its insurance rates. As a result, many carriers are eager for information that will help them predict what drivers are most likely to get into a crash, and what factors will most likely lead to a crash.
Shippers are also eager to work with safe carriers, as a reduction in crash rates results in fewer cargo incidents.
Software companies are answering the call. Although predictive risk modeling has been around for years, advances in artificial intelligence (AI) and machine learning are helping technologists put a finer point on that information, as new programs ingest data from a wide variety of sources to help identity the safest drivers and carriers.
Among those leading the charge in AI safety innovation is the digital freight network Convoy, where a 16-person team sets safety and performance standards for carriers, then uses data-driven processes to manage those standards in real-time.
Last year, the company reported on the performance of an in-house compliance program that used the Federal Motor Carrier Safety Administration’s BASICs and Safety Ratings data to vet carriers on its network.
“While that standard was more scrupulous than any other we’d seen at the time,” Lorin Seeks, Convoy’s director of carrier quality and compliance, told FreightWaves, “we still saw opportunities to further improve.” Simultaneously, Seeks said, Convoy was testing a new machine learning model that uses significantly more granular data.
As of today, it’s out with the old and in with the new. Seattle-based Convoy has announced that it is rolling out a novel application of machine learning and automation that identifies the safest carriers to allow into its network. The program yields 16% fewer accidents than the industry average, the company says.
Additionally, Convoy’s claim rate clocks in at less than once per 2,000 loads, whereas the industry experiences a cargo claim about once per 100 loads.
According to Seeks, the widely used FMCSA data generates overall ratings for carriers. But 95% of carriers have not completed a full compliance review and remain “not rated,” limiting visibility into the vast majority of safety records and preventing adequate assessment of carrier safety.
Convoy’s system applies machine learning to more accurately predict which carriers are likely to get into accidents by processing thousands of inputs – such as carrier crash history, vehicle maintenance, and speeding and traffic violations – across millions of records.
The algorithm then produces scores for the carriers in Convoy’s network. It automatically reviews those scores, qualifying or removing carriers, in real-time, from the network that fall below safe thresholds.
The model also gets smarter over time, driving continuous improvements as more data is generated, providing shippers with increasingly high levels of safe and reliable carriers.
In recent years there have been “enormous advances” in big data analytics that have improved transportation predictive modeling, Seeks said.
These include the application of machine learning techniques that are able to detect new patterns without human intervention, and the ability to model risk using real-time sensor data, which until recently was “prohibitively computationally intensive.”
But prior to Convoy, Seeks claimed, “we’re not aware of any other transportation intermediaries that had the in-house technical sophistication to apply them.”
Convoy’s network focuses primarily on smaller carriers, which typically have higher crash rates per truck than larger carriers. “But part of the beauty of Convoy’s new model is that these group differences become irrelevant,” Seeks said. “Our model is able to select the safest carriers regardless of their size and qualify them.”
If motor carriers improve their scores and are once again above the Convoy threshold, they are automatically reactivated into the network, assuming they meet all other eligibility criteria, according to Seeks.
Click here to read more FreightWaves articles by Linda Baker.