The views expressed here are solely those of the author and do not necessarily represent the views of FreightWaves or its affiliates.
In this installment of the AI in Supply Chain series (#AIinSupplyChain) we explore how AI is being used to help freight forwarders and customs brokers manage their workflows more efficiently.
To help us understand the issues at play, we will meet Vector.ai, a startup based in London. Regular readers of this column have previously encountered Vector.ai in Commentary: How supply chain startups are surviving COVID-19.
This article is most directly related to the numerous past articles in this series in which we have touched on the fractious dichotomy between digital freight forwarders and their incumbent counterparts. For example Commentary: Digital freight brokers face a moment of truth and Commentary: Why are digital freight brokers struggling to solve their Uber-scale problems?, among others.
The communication problem in logistics
In every conversation I have had with professionals in supply chain logistics, the ubiquity of paper-based workflows and business processes inevitably comes up.
The magnitude of the problem is captured vividly in J.P.Morgan’s 2017 Trade Outlook: “With export and import trade shipments requiring an average of 36 original documents, 240 copies and the involvement of 27 entities, Fortune 500 companies incur more than $81 billion of unnecessary supply chain and working capital costs each year due to inefficiencies and lack of visibility.”
That estimate applies to Fortune 500 companies only. So, it is obviously only the tip of the iceberg; one expects that Fortune 500 companies can afford to invest in state-of-the-art technology to streamline and optimize their communications workflows and processes, while smaller more resource-constrained companies must live with less effective tools.
This is the problem that companies like Vector.ai set out to solve.
Insights from the field — Vector.ai
James Coombes is CEO and co-founder of Vector.ai. I asked him to tell me more about the problem Vector.ai solves for its customers.
He said, “The days of freight forwarders and customs brokers managing their customer workflow with email and document keying in are numbered.”
He continues, “It isn’t a new ambition, we’ve just lacked the right tools until now. Previously the focus was uniquely on technology like EDI and APIs to drive digitization. But now the market has accepted that there is no one-size-fits-all solution. This is a human challenge that can only be solved with intelligent tools, not dumb ones.”
This is where Vector.ai comes in.
According to Coombes, Vector.ai has built a unique platform that is robust enough to understand the sorts of email content and documents that freight forwarders and customs brokers encounter in high volume every day.
He says Vector.ai empowers operators by intelligently gathering the information they need from a given transaction, whether it is in document or email, and allows them to rapidly respond on the back of that data. This leads to quicker decisions, quicker turnaround times and happier customers.
Understanding the technology challenge
I asked Coombes to tell me a bit about Vector.ai’s secret sauce, the unique insight or approach that is helping Vector.ai’s team find success with customers in a market that can be skeptical about adopting new technology.
“Logistics communication data is an exciting challenge that is really well suited to AI,” he said. “The content of these documents and emails changes, but within a well-defined context; we’re not asking the platform to figure out what to do with that invite to your best friend’s birthday party, for instance. In logistics there’s a large but fortunately limited distribution of data points and actions that need to be taken on any transaction, e.g., a set of documents on a pre-alert notice. That constraint makes this a great problem for machine learning-based technology.”
However, Coombes points out that although the problem may be well defined, that does not imply that the technology is easy to build or simple to complement in the real world.
He says, “We’re very product-focused; you have to be. But behind the scenes, the technology required to build this is astonishingly hard.”
What’s under the hood of Vector.ai’s product?
As I have been doing with everyone I have covered for this #AIinSupplyChain series, I asked Coombes: “What is unique about your approach? Deep learning seems to be all the rage these days? Does Vector.AI use a form of deep learning? Reinforcement learning?”
He responded, “Our machine learning models are built to work across a spectrum of data availability, from low- and poor-quality data right through to rich datasets, and we’ve pioneered approaches across various techniques: from deep learning to reinforcement learning and more, to maintain performance across the different data distributions our customers have at different times. This is an ongoing research focus of ours. Drawing on expertise in computer vision and natural language processing, we actively use state-of-the-art techniques from the academic community that are barely a few months old and make them real and accessible to our customers.”
Commenting on what Vector.ai has accomplished, Roberto Cipolla, a professor of information engineering at the University of Cambridge, who is also a technical adviser to Vector.ai, said, “Within the trade problem set, Vector.ai’s core technology has succeeded where others have fallen well short. The level of complexity involved in digitizing fragmented, noisy and even free-form information within this sector cannot be understated.”
On selling AI and machine learning to a legacy industry with digital immigrant customers
Finally, we got to talking about what the team at Vector.ai is learning about selling AI and machine learning software to customers who are digital immigrants, in legacy industries.
According to Coombes, one pitfall is startups or companies that advertise AI and machine learning, but actually do something else.
“We see a lot of providers talking about AI but few who actually do it,” he said. “What hurts the market is the fake-it-till-you-make-it approach, where companies have embraced the idea of doing unsustainable things to scale, like secretly having hundreds of employees behind the scenes ‘just for now.’ It makes companies good at offshoring things but not very good at building out a scalable AI strategy.”
In my own experience since 2008, I have encountered the problem Coombes is describing. Moreover, the problem is covered in the press. For example, this July 20 story in Forbes, ScaleFactor Raised $100 Million In A Year Then Blamed Covid-19 For Its Demise. Employees Say It Had Much Bigger Problems, describes how investors were fooled.
Describing another pitfall that startups like Vector.ai might encounter as they seek to grow, Coombes also talks about what he terms “lazy sales.”
He says, “Another thing that doesn’t work is what we call lazy sales. It is the easiest job in the world to pitch an AI platform that is 98-100% accurate. The harder job is to tell your customers that sometimes machine learning gets things wrong, but the platform is designed to help operators mitigate exceptions in a robust way, which can drive future accuracy gains. Some of our best customers have been sold the dream, gone through years of disappointment and then come to us. We don’t sugarcoat things.”
How to separate genuine AI and machine learning startups from everyone else
I am assuming that most FreightWaves readers who read this column are engaging with AI and machine learning in a consistent way for the first time in this #AIinSupplyChain series, so I asked Coombes what he suggests one might do in order to separate startups that are doing genuine AI and machine learning work that has a chance of successfully solving customers’ problems in the real world from startups that are mainly selling empty promises.
He said, “One of the best ways to differentiate between those who say they do and those that just actually do, is to look closely at the team.”
He added, “We’ve been building Vector.ai since 2017. Half of our founding team and half of our engineering team, which includes two current machine learning professors at Cambridge University and UCL [University College London], is dedicated to the machine learning and artificial intelligence aspects of solving this problem. That’s the kind of talent you need to do something like this properly.”
Given the many challenges that have been exacerbated by COVID19, I believe that the issues around supply chain workflow automation will only become more acute over time. That creates a promising opportunity for productivity-boosting startups like Vector.ai as businesses seek to modernize their business process flows to more closely align with the ways in which people now collaborate with one another within and across company boundaries and national borders.
If you are a team working on innovations that you believe have the potential to significantly refashion global supply chains, we’d love to tell your story in FreightWaves. I am easy to reach on LinkedIn and Twitter. Alternatively, you can reach out to any member of the editorial team at FreightWaves at [email protected]
Dig deeper into the #AIinSupplyChain Series with FreightWaves
Author’s disclosure: I am not an investor in any early-stage startups mentioned in this article, either personally or through REFASHIOND Ventures. I have no other financial relationship with any entities mentioned in this article.