Cargo’s Crystal Ball: The power of machine learning
Most shippers were caught off-guard last August when the world’s seventh-largest container line suddenly went into receivership. An estimated US$14 billion in cargo was left in limbo after two-thirds of South Korea-based Hanjin’s fleet – some 93 vessels – were seized, stranded or denied entry into ports. Shippers immediately opened their contingency playbooks and began the costly process of rebooking shipments on other container lines; in some cases, shippers had to resort to alternative modes of transportation to mitigate shipment delays.
But what if shippers had known well in advance that Hanjin was on the brink of collapse? With the diffusion of so-called “machine learning” technologies, such foresight into looming supply-chain disruptions may no longer be the undertaking of a soothsayer with a crystal ball.
Although early warning systems have long been around to help cargo avoid weather hazards, such as an oncoming blizzard or hurricane, utilizing available data to detect the deteriorating financial state of a supply chain partner requires a different barometer. DHL Resilience 360’s Supply Watch, for instance, utilizes a machine-learning algorithm to recognize indicators for more than 140 different types of risk. Supply Watch was created out of a desire to pick up the “more nuanced risks, like supply failure, financial issues and compliance issues that may be very specific to companies and suppliers that our customers work with,” said Shehrina Kamal, senior product manager at DHL Resilience 360.
Machine learning represents the next iteration of data analytics, and finds applications in situations where rules-based programs cannot be developed to determine a solution, or when the volume of data is too large to handle manually. The algorithm “gives us an automated way of dealing with vast amounts of data in a proactive way,” said Kamal.
Supply Watch parses upwards of 30 million posts from 300,000 online and social media sources to find very subtle indicators – the needles in haystacks of information. While manual analysis could pick up on some of these risks, the predictive machine-learning algorithm is constantly evolving, and can be “taught” to seek out risk indicators as they relate to a particular company.
Supply Watch was launched after Hanjin’s collapse, but the algorithm that powers it has been in development for more than 18 months. “When we look back at all the data we were collecting, we realized that, in the case of Hanjin, there were indications that it might go into receivership” as early as March 2016, said Kamal. At this point, forwarders or shippers linked to Hanjin would have been notified of the mounting risk, and could have begun evaluating alternative options.
With the proliferation of “Big Data” throughout the logistics industry, supply chain early warning systems are just one of many ways machine learning technologies are being utilized to improve supply chain efficiency and visibility, while giving forwarders and carriers an edge in meeting shippers’ demands. Other examples include and optimizing freight placement as it moves through the supply chain, enhancing arrival time estimates, and the forecasting and calculation of spot rates.
