Forwarding is a document-intensive business, but it’s only recently that global forwarders have gotten a true sense of the scale of inefficiencies tied to the lack of structure inherent in core documents such as the bill of lading, commercial invoices, and packing list.
In the absence of a true paperless environment, where the documents would be generated and conveyed digitally, forwarders are generally left with only a few options. They can manually enter data from documents into their systems of record, they can use general purpose optical character recognition (OCR) software to convert those documents into a digital format, or they can forsake keeping digital storehouses of the data contained in key shipping documents.
Given that scenario, London-based technology startup Vector.ai, founded in 2017, has focused on providing forwarders with a more effective solution to turn the information captive in documents into usable data. Vector.ai landed a seed round in December, CEO James Coombes told JOC.com, and has been building out its team since.
Coombes, who has a background in finance, initially set out to build a machine learning algorithm-based product to tackle the complicated issue of trade finance. And while Vector.ai still helps banks automate decisions on which companies to finance, Coombes realized, after piloting a project with a global forwarder, that there was an adjacent problem just as ticklish.
“As a forwarder, margins are tight, and so they should all be looking to increase the revenue they make on a transaction or reduce the cost to serve the customer,” he said. “Nobody in any supply chain has any control over what data anyone else inputs. They try, but a forwarder cannot dictate that this is the format you need to provide. So we’re like a universal adapter for them.”
A familiar problem
Coombes said the problem forwarders face in terms of managing documentation was similar to one faced by banks — regulation. “They both have regulatory and sanctions types of requirements,” he said. “We built the [machine learning] models for the docs that the banks need, so we turned to focus on the trade compliance docs that forwarders manage.”
At the heart of the Vector.ai solution is a machine learning algorithm that, coupled with OCR technology, learns over time how to accurately read data in those core shipping documents, no matter what format or how illegible the information is. The program is set up to learn more over time, so the more documents it reads, the better it becomes at understanding what it is reading.
The benefits to forwarders are operational efficiencies, better data accuracy, and the opportunity to leverage the aggregate data into better serving their shipper customers, Coombes said.
“Forwarders have solved the back office and the customer-facing side, but haven’t solved the operational side,” he said, referring to technology that supports operations. “If you can understand the operational layer, you can understand other things. It helps to know your transactions, to know your customer, and then you can pipe things in like insurance and trade finance.”
Vector.ai has been piloting its solution with a major global forwarder for the last year. A senior executive with the forwarder, who did not want to be identified, told JOC.com that Vector.ai’s product has been significantly more accurate in helping it convert documents into data than other homegrown or off-the-shelf OCR approaches the forwarder has previously tried.
“Confidence in the accuracy of those solutions was good, but not great,” the source said. “The trouble is, when you’re dealing with trade compliance documents, good is not enough.”
The forwarder executive said third-party logistics providers that manage freight and customs compliance are the ideal testing ground for the Vector.ai machine learning algorithm.
“A [customs] broker and forwarder, you have all these files from previous shipments,” he said. “There’s no better testing environment for machine learning, because you just learn millions of times over and you’re not affecting any ongoing shipments. As you provide information, there’s a feedback loop where it teaches itself. We put that into tests, and the early results are remarkable.”
A host of technology providers have sprung up in recent years aiming to use machine learning and process automation to help forwarders improve processes such as customer service, data extraction from existing systems, and problem resolution.
Contact Eric Johnson at [email protected] and follow him on Twitter: @LogTechEric.
Recent Comments